SemaChecking.cpp revision f65076ebf5e8bc5ae7b77fe3c3a65dc5db8d763f
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  return false;
322}
323
324bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) {
325  // Printf checking.
326  const FormatAttr *Format = NDecl->getAttr<FormatAttr>();
327  if (!Format)
328    return false;
329
330  const VarDecl *V = dyn_cast<VarDecl>(NDecl);
331  if (!V)
332    return false;
333
334  QualType Ty = V->getType();
335  if (!Ty->isBlockPointerType())
336    return false;
337
338  const bool b = Format->getType() == "scanf";
339  if (!b && !CheckablePrintfAttr(Format, TheCall))
340    return false;
341
342  bool HasVAListArg = Format->getFirstArg() == 0;
343  CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1,
344                            HasVAListArg ? 0 : Format->getFirstArg() - 1, !b);
345
346  return false;
347}
348
349/// SemaBuiltinAtomicOverloaded - We have a call to a function like
350/// __sync_fetch_and_add, which is an overloaded function based on the pointer
351/// type of its first argument.  The main ActOnCallExpr routines have already
352/// promoted the types of arguments because all of these calls are prototyped as
353/// void(...).
354///
355/// This function goes through and does final semantic checking for these
356/// builtins,
357ExprResult
358Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) {
359  CallExpr *TheCall = (CallExpr *)TheCallResult.get();
360  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
361  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
362
363  // Ensure that we have at least one argument to do type inference from.
364  if (TheCall->getNumArgs() < 1) {
365    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
366      << 0 << 1 << TheCall->getNumArgs()
367      << TheCall->getCallee()->getSourceRange();
368    return ExprError();
369  }
370
371  // Inspect the first argument of the atomic builtin.  This should always be
372  // a pointer type, whose element is an integral scalar or pointer type.
373  // Because it is a pointer type, we don't have to worry about any implicit
374  // casts here.
375  // FIXME: We don't allow floating point scalars as input.
376  Expr *FirstArg = TheCall->getArg(0);
377  if (!FirstArg->getType()->isPointerType()) {
378    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
379      << FirstArg->getType() << FirstArg->getSourceRange();
380    return ExprError();
381  }
382
383  QualType ValType =
384    FirstArg->getType()->getAs<PointerType>()->getPointeeType();
385  if (!ValType->isIntegerType() && !ValType->isAnyPointerType() &&
386      !ValType->isBlockPointerType()) {
387    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr)
388      << FirstArg->getType() << FirstArg->getSourceRange();
389    return ExprError();
390  }
391
392  // The majority of builtins return a value, but a few have special return
393  // types, so allow them to override appropriately below.
394  QualType ResultType = ValType;
395
396  // We need to figure out which concrete builtin this maps onto.  For example,
397  // __sync_fetch_and_add with a 2 byte object turns into
398  // __sync_fetch_and_add_2.
399#define BUILTIN_ROW(x) \
400  { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
401    Builtin::BI##x##_8, Builtin::BI##x##_16 }
402
403  static const unsigned BuiltinIndices[][5] = {
404    BUILTIN_ROW(__sync_fetch_and_add),
405    BUILTIN_ROW(__sync_fetch_and_sub),
406    BUILTIN_ROW(__sync_fetch_and_or),
407    BUILTIN_ROW(__sync_fetch_and_and),
408    BUILTIN_ROW(__sync_fetch_and_xor),
409
410    BUILTIN_ROW(__sync_add_and_fetch),
411    BUILTIN_ROW(__sync_sub_and_fetch),
412    BUILTIN_ROW(__sync_and_and_fetch),
413    BUILTIN_ROW(__sync_or_and_fetch),
414    BUILTIN_ROW(__sync_xor_and_fetch),
415
416    BUILTIN_ROW(__sync_val_compare_and_swap),
417    BUILTIN_ROW(__sync_bool_compare_and_swap),
418    BUILTIN_ROW(__sync_lock_test_and_set),
419    BUILTIN_ROW(__sync_lock_release),
420    BUILTIN_ROW(__sync_swap)
421  };
422#undef BUILTIN_ROW
423
424  // Determine the index of the size.
425  unsigned SizeIndex;
426  switch (Context.getTypeSizeInChars(ValType).getQuantity()) {
427  case 1: SizeIndex = 0; break;
428  case 2: SizeIndex = 1; break;
429  case 4: SizeIndex = 2; break;
430  case 8: SizeIndex = 3; break;
431  case 16: SizeIndex = 4; break;
432  default:
433    Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
434      << FirstArg->getType() << FirstArg->getSourceRange();
435    return ExprError();
436  }
437
438  // Each of these builtins has one pointer argument, followed by some number of
439  // values (0, 1 or 2) followed by a potentially empty varags list of stuff
440  // that we ignore.  Find out which row of BuiltinIndices to read from as well
441  // as the number of fixed args.
442  unsigned BuiltinID = FDecl->getBuiltinID();
443  unsigned BuiltinIndex, NumFixed = 1;
444  switch (BuiltinID) {
445  default: assert(0 && "Unknown overloaded atomic builtin!");
446  case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break;
447  case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break;
448  case Builtin::BI__sync_fetch_and_or:  BuiltinIndex = 2; break;
449  case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break;
450  case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break;
451
452  case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break;
453  case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break;
454  case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break;
455  case Builtin::BI__sync_or_and_fetch:  BuiltinIndex = 8; break;
456  case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break;
457
458  case Builtin::BI__sync_val_compare_and_swap:
459    BuiltinIndex = 10;
460    NumFixed = 2;
461    break;
462  case Builtin::BI__sync_bool_compare_and_swap:
463    BuiltinIndex = 11;
464    NumFixed = 2;
465    ResultType = Context.BoolTy;
466    break;
467  case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break;
468  case Builtin::BI__sync_lock_release:
469    BuiltinIndex = 13;
470    NumFixed = 0;
471    ResultType = Context.VoidTy;
472    break;
473  case Builtin::BI__sync_swap: BuiltinIndex = 14; break;
474  }
475
476  // Now that we know how many fixed arguments we expect, first check that we
477  // have at least that many.
478  if (TheCall->getNumArgs() < 1+NumFixed) {
479    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
480      << 0 << 1+NumFixed << TheCall->getNumArgs()
481      << TheCall->getCallee()->getSourceRange();
482    return ExprError();
483  }
484
485  // Get the decl for the concrete builtin from this, we can tell what the
486  // concrete integer type we should convert to is.
487  unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
488  const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
489  IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName);
490  FunctionDecl *NewBuiltinDecl =
491    cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID,
492                                           TUScope, false, DRE->getLocStart()));
493
494  // The first argument --- the pointer --- has a fixed type; we
495  // deduce the types of the rest of the arguments accordingly.  Walk
496  // the remaining arguments, converting them to the deduced value type.
497  for (unsigned i = 0; i != NumFixed; ++i) {
498    ExprResult Arg = TheCall->getArg(i+1);
499
500    // If the argument is an implicit cast, then there was a promotion due to
501    // "...", just remove it now.
502    if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg.get())) {
503      Arg = ICE->getSubExpr();
504      ICE->setSubExpr(0);
505      TheCall->setArg(i+1, Arg.get());
506    }
507
508    // GCC does an implicit conversion to the pointer or integer ValType.  This
509    // can fail in some cases (1i -> int**), check for this error case now.
510    CastKind Kind = CK_Invalid;
511    ExprValueKind VK = VK_RValue;
512    CXXCastPath BasePath;
513    Arg = CheckCastTypes(Arg.get()->getSourceRange(), ValType, Arg.take(), Kind, VK, BasePath);
514    if (Arg.isInvalid())
515      return ExprError();
516
517    // Okay, we have something that *can* be converted to the right type.  Check
518    // to see if there is a potentially weird extension going on here.  This can
519    // happen when you do an atomic operation on something like an char* and
520    // pass in 42.  The 42 gets converted to char.  This is even more strange
521    // for things like 45.123 -> char, etc.
522    // FIXME: Do this check.
523    Arg = ImpCastExprToType(Arg.take(), ValType, Kind, VK, &BasePath);
524    TheCall->setArg(i+1, Arg.get());
525  }
526
527  // Switch the DeclRefExpr to refer to the new decl.
528  DRE->setDecl(NewBuiltinDecl);
529  DRE->setType(NewBuiltinDecl->getType());
530
531  // Set the callee in the CallExpr.
532  // FIXME: This leaks the original parens and implicit casts.
533  ExprResult PromotedCall = UsualUnaryConversions(DRE);
534  if (PromotedCall.isInvalid())
535    return ExprError();
536  TheCall->setCallee(PromotedCall.take());
537
538  // Change the result type of the call to match the original value type. This
539  // is arbitrary, but the codegen for these builtins ins design to handle it
540  // gracefully.
541  TheCall->setType(ResultType);
542
543  return move(TheCallResult);
544}
545
546
547/// CheckObjCString - Checks that the argument to the builtin
548/// CFString constructor is correct
549/// Note: It might also make sense to do the UTF-16 conversion here (would
550/// simplify the backend).
551bool Sema::CheckObjCString(Expr *Arg) {
552  Arg = Arg->IgnoreParenCasts();
553  StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
554
555  if (!Literal || Literal->isWide()) {
556    Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
557      << Arg->getSourceRange();
558    return true;
559  }
560
561  if (Literal->containsNonAsciiOrNull()) {
562    llvm::StringRef String = Literal->getString();
563    unsigned NumBytes = String.size();
564    llvm::SmallVector<UTF16, 128> ToBuf(NumBytes);
565    const UTF8 *FromPtr = (UTF8 *)String.data();
566    UTF16 *ToPtr = &ToBuf[0];
567
568    ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes,
569                                                 &ToPtr, ToPtr + NumBytes,
570                                                 strictConversion);
571    // Check for conversion failure.
572    if (Result != conversionOK)
573      Diag(Arg->getLocStart(),
574           diag::warn_cfstring_truncated) << Arg->getSourceRange();
575  }
576  return false;
577}
578
579/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
580/// Emit an error and return true on failure, return false on success.
581bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
582  Expr *Fn = TheCall->getCallee();
583  if (TheCall->getNumArgs() > 2) {
584    Diag(TheCall->getArg(2)->getLocStart(),
585         diag::err_typecheck_call_too_many_args)
586      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
587      << Fn->getSourceRange()
588      << SourceRange(TheCall->getArg(2)->getLocStart(),
589                     (*(TheCall->arg_end()-1))->getLocEnd());
590    return true;
591  }
592
593  if (TheCall->getNumArgs() < 2) {
594    return Diag(TheCall->getLocEnd(),
595      diag::err_typecheck_call_too_few_args_at_least)
596      << 0 /*function call*/ << 2 << TheCall->getNumArgs();
597  }
598
599  // Determine whether the current function is variadic or not.
600  BlockScopeInfo *CurBlock = getCurBlock();
601  bool isVariadic;
602  if (CurBlock)
603    isVariadic = CurBlock->TheDecl->isVariadic();
604  else if (FunctionDecl *FD = getCurFunctionDecl())
605    isVariadic = FD->isVariadic();
606  else
607    isVariadic = getCurMethodDecl()->isVariadic();
608
609  if (!isVariadic) {
610    Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
611    return true;
612  }
613
614  // Verify that the second argument to the builtin is the last argument of the
615  // current function or method.
616  bool SecondArgIsLastNamedArgument = false;
617  const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
618
619  if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
620    if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
621      // FIXME: This isn't correct for methods (results in bogus warning).
622      // Get the last formal in the current function.
623      const ParmVarDecl *LastArg;
624      if (CurBlock)
625        LastArg = *(CurBlock->TheDecl->param_end()-1);
626      else if (FunctionDecl *FD = getCurFunctionDecl())
627        LastArg = *(FD->param_end()-1);
628      else
629        LastArg = *(getCurMethodDecl()->param_end()-1);
630      SecondArgIsLastNamedArgument = PV == LastArg;
631    }
632  }
633
634  if (!SecondArgIsLastNamedArgument)
635    Diag(TheCall->getArg(1)->getLocStart(),
636         diag::warn_second_parameter_of_va_start_not_last_named_argument);
637  return false;
638}
639
640/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
641/// friends.  This is declared to take (...), so we have to check everything.
642bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
643  if (TheCall->getNumArgs() < 2)
644    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
645      << 0 << 2 << TheCall->getNumArgs()/*function call*/;
646  if (TheCall->getNumArgs() > 2)
647    return Diag(TheCall->getArg(2)->getLocStart(),
648                diag::err_typecheck_call_too_many_args)
649      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
650      << SourceRange(TheCall->getArg(2)->getLocStart(),
651                     (*(TheCall->arg_end()-1))->getLocEnd());
652
653  ExprResult OrigArg0 = TheCall->getArg(0);
654  ExprResult OrigArg1 = TheCall->getArg(1);
655
656  // Do standard promotions between the two arguments, returning their common
657  // type.
658  QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
659  if (OrigArg0.isInvalid() || OrigArg1.isInvalid())
660    return true;
661
662  // Make sure any conversions are pushed back into the call; this is
663  // type safe since unordered compare builtins are declared as "_Bool
664  // foo(...)".
665  TheCall->setArg(0, OrigArg0.get());
666  TheCall->setArg(1, OrigArg1.get());
667
668  if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent())
669    return false;
670
671  // If the common type isn't a real floating type, then the arguments were
672  // invalid for this operation.
673  if (!Res->isRealFloatingType())
674    return Diag(OrigArg0.get()->getLocStart(),
675                diag::err_typecheck_call_invalid_ordered_compare)
676      << OrigArg0.get()->getType() << OrigArg1.get()->getType()
677      << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd());
678
679  return false;
680}
681
682/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like
683/// __builtin_isnan and friends.  This is declared to take (...), so we have
684/// to check everything. We expect the last argument to be a floating point
685/// value.
686bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) {
687  if (TheCall->getNumArgs() < NumArgs)
688    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
689      << 0 << NumArgs << TheCall->getNumArgs()/*function call*/;
690  if (TheCall->getNumArgs() > NumArgs)
691    return Diag(TheCall->getArg(NumArgs)->getLocStart(),
692                diag::err_typecheck_call_too_many_args)
693      << 0 /*function call*/ << NumArgs << TheCall->getNumArgs()
694      << SourceRange(TheCall->getArg(NumArgs)->getLocStart(),
695                     (*(TheCall->arg_end()-1))->getLocEnd());
696
697  Expr *OrigArg = TheCall->getArg(NumArgs-1);
698
699  if (OrigArg->isTypeDependent())
700    return false;
701
702  // This operation requires a non-_Complex floating-point number.
703  if (!OrigArg->getType()->isRealFloatingType())
704    return Diag(OrigArg->getLocStart(),
705                diag::err_typecheck_call_invalid_unary_fp)
706      << OrigArg->getType() << OrigArg->getSourceRange();
707
708  // If this is an implicit conversion from float -> double, remove it.
709  if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) {
710    Expr *CastArg = Cast->getSubExpr();
711    if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) {
712      assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) &&
713             "promotion from float to double is the only expected cast here");
714      Cast->setSubExpr(0);
715      TheCall->setArg(NumArgs-1, CastArg);
716      OrigArg = CastArg;
717    }
718  }
719
720  return false;
721}
722
723/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
724// This is declared to take (...), so we have to check everything.
725ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
726  if (TheCall->getNumArgs() < 2)
727    return ExprError(Diag(TheCall->getLocEnd(),
728                          diag::err_typecheck_call_too_few_args_at_least)
729      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
730      << TheCall->getSourceRange());
731
732  // Determine which of the following types of shufflevector we're checking:
733  // 1) unary, vector mask: (lhs, mask)
734  // 2) binary, vector mask: (lhs, rhs, mask)
735  // 3) binary, scalar mask: (lhs, rhs, index, ..., index)
736  QualType resType = TheCall->getArg(0)->getType();
737  unsigned numElements = 0;
738
739  if (!TheCall->getArg(0)->isTypeDependent() &&
740      !TheCall->getArg(1)->isTypeDependent()) {
741    QualType LHSType = TheCall->getArg(0)->getType();
742    QualType RHSType = TheCall->getArg(1)->getType();
743
744    if (!LHSType->isVectorType() || !RHSType->isVectorType()) {
745      Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
746        << SourceRange(TheCall->getArg(0)->getLocStart(),
747                       TheCall->getArg(1)->getLocEnd());
748      return ExprError();
749    }
750
751    numElements = LHSType->getAs<VectorType>()->getNumElements();
752    unsigned numResElements = TheCall->getNumArgs() - 2;
753
754    // Check to see if we have a call with 2 vector arguments, the unary shuffle
755    // with mask.  If so, verify that RHS is an integer vector type with the
756    // same number of elts as lhs.
757    if (TheCall->getNumArgs() == 2) {
758      if (!RHSType->hasIntegerRepresentation() ||
759          RHSType->getAs<VectorType>()->getNumElements() != numElements)
760        Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
761          << SourceRange(TheCall->getArg(1)->getLocStart(),
762                         TheCall->getArg(1)->getLocEnd());
763      numResElements = numElements;
764    }
765    else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) {
766      Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
767        << SourceRange(TheCall->getArg(0)->getLocStart(),
768                       TheCall->getArg(1)->getLocEnd());
769      return ExprError();
770    } else if (numElements != numResElements) {
771      QualType eltType = LHSType->getAs<VectorType>()->getElementType();
772      resType = Context.getVectorType(eltType, numResElements,
773                                      VectorType::GenericVector);
774    }
775  }
776
777  for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
778    if (TheCall->getArg(i)->isTypeDependent() ||
779        TheCall->getArg(i)->isValueDependent())
780      continue;
781
782    llvm::APSInt Result(32);
783    if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
784      return ExprError(Diag(TheCall->getLocStart(),
785                  diag::err_shufflevector_nonconstant_argument)
786                << TheCall->getArg(i)->getSourceRange());
787
788    if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
789      return ExprError(Diag(TheCall->getLocStart(),
790                  diag::err_shufflevector_argument_too_large)
791               << TheCall->getArg(i)->getSourceRange());
792  }
793
794  llvm::SmallVector<Expr*, 32> exprs;
795
796  for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
797    exprs.push_back(TheCall->getArg(i));
798    TheCall->setArg(i, 0);
799  }
800
801  return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(),
802                                            exprs.size(), resType,
803                                            TheCall->getCallee()->getLocStart(),
804                                            TheCall->getRParenLoc()));
805}
806
807/// SemaBuiltinPrefetch - Handle __builtin_prefetch.
808// This is declared to take (const void*, ...) and can take two
809// optional constant int args.
810bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
811  unsigned NumArgs = TheCall->getNumArgs();
812
813  if (NumArgs > 3)
814    return Diag(TheCall->getLocEnd(),
815             diag::err_typecheck_call_too_many_args_at_most)
816             << 0 /*function call*/ << 3 << NumArgs
817             << TheCall->getSourceRange();
818
819  // Argument 0 is checked for us and the remaining arguments must be
820  // constant integers.
821  for (unsigned i = 1; i != NumArgs; ++i) {
822    Expr *Arg = TheCall->getArg(i);
823
824    llvm::APSInt Result;
825    if (SemaBuiltinConstantArg(TheCall, i, Result))
826      return true;
827
828    // FIXME: gcc issues a warning and rewrites these to 0. These
829    // seems especially odd for the third argument since the default
830    // is 3.
831    if (i == 1) {
832      if (Result.getLimitedValue() > 1)
833        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
834             << "0" << "1" << Arg->getSourceRange();
835    } else {
836      if (Result.getLimitedValue() > 3)
837        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
838            << "0" << "3" << Arg->getSourceRange();
839    }
840  }
841
842  return false;
843}
844
845/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr
846/// TheCall is a constant expression.
847bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,
848                                  llvm::APSInt &Result) {
849  Expr *Arg = TheCall->getArg(ArgNum);
850  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
851  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
852
853  if (Arg->isTypeDependent() || Arg->isValueDependent()) return false;
854
855  if (!Arg->isIntegerConstantExpr(Result, Context))
856    return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type)
857                << FDecl->getDeclName() <<  Arg->getSourceRange();
858
859  return false;
860}
861
862/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
863/// int type). This simply type checks that type is one of the defined
864/// constants (0-3).
865// For compatability check 0-3, llvm only handles 0 and 2.
866bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
867  llvm::APSInt Result;
868
869  // Check constant-ness first.
870  if (SemaBuiltinConstantArg(TheCall, 1, Result))
871    return true;
872
873  Expr *Arg = TheCall->getArg(1);
874  if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
875    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
876             << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
877  }
878
879  return false;
880}
881
882/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
883/// This checks that val is a constant 1.
884bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
885  Expr *Arg = TheCall->getArg(1);
886  llvm::APSInt Result;
887
888  // TODO: This is less than ideal. Overload this to take a value.
889  if (SemaBuiltinConstantArg(TheCall, 1, Result))
890    return true;
891
892  if (Result != 1)
893    return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
894             << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
895
896  return false;
897}
898
899// Handle i > 1 ? "x" : "y", recursively.
900bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall,
901                                  bool HasVAListArg,
902                                  unsigned format_idx, unsigned firstDataArg,
903                                  bool isPrintf) {
904 tryAgain:
905  if (E->isTypeDependent() || E->isValueDependent())
906    return false;
907
908  switch (E->getStmtClass()) {
909  case Stmt::BinaryConditionalOperatorClass:
910  case Stmt::ConditionalOperatorClass: {
911    const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E);
912    return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg,
913                                  format_idx, firstDataArg, isPrintf)
914        && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg,
915                                  format_idx, firstDataArg, isPrintf);
916  }
917
918  case Stmt::IntegerLiteralClass:
919    // Technically -Wformat-nonliteral does not warn about this case.
920    // The behavior of printf and friends in this case is implementation
921    // dependent.  Ideally if the format string cannot be null then
922    // it should have a 'nonnull' attribute in the function prototype.
923    return true;
924
925  case Stmt::ImplicitCastExprClass: {
926    E = cast<ImplicitCastExpr>(E)->getSubExpr();
927    goto tryAgain;
928  }
929
930  case Stmt::ParenExprClass: {
931    E = cast<ParenExpr>(E)->getSubExpr();
932    goto tryAgain;
933  }
934
935  case Stmt::OpaqueValueExprClass:
936    if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) {
937      E = src;
938      goto tryAgain;
939    }
940    return false;
941
942  case Stmt::PredefinedExprClass:
943    // While __func__, etc., are technically not string literals, they
944    // cannot contain format specifiers and thus are not a security
945    // liability.
946    return true;
947
948  case Stmt::DeclRefExprClass: {
949    const DeclRefExpr *DR = cast<DeclRefExpr>(E);
950
951    // As an exception, do not flag errors for variables binding to
952    // const string literals.
953    if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
954      bool isConstant = false;
955      QualType T = DR->getType();
956
957      if (const ArrayType *AT = Context.getAsArrayType(T)) {
958        isConstant = AT->getElementType().isConstant(Context);
959      } else if (const PointerType *PT = T->getAs<PointerType>()) {
960        isConstant = T.isConstant(Context) &&
961                     PT->getPointeeType().isConstant(Context);
962      }
963
964      if (isConstant) {
965        if (const Expr *Init = VD->getAnyInitializer())
966          return SemaCheckStringLiteral(Init, TheCall,
967                                        HasVAListArg, format_idx, firstDataArg,
968                                        isPrintf);
969      }
970
971      // For vprintf* functions (i.e., HasVAListArg==true), we add a
972      // special check to see if the format string is a function parameter
973      // of the function calling the printf function.  If the function
974      // has an attribute indicating it is a printf-like function, then we
975      // should suppress warnings concerning non-literals being used in a call
976      // to a vprintf function.  For example:
977      //
978      // void
979      // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
980      //      va_list ap;
981      //      va_start(ap, fmt);
982      //      vprintf(fmt, ap);  // Do NOT emit a warning about "fmt".
983      //      ...
984      //
985      //
986      //  FIXME: We don't have full attribute support yet, so just check to see
987      //    if the argument is a DeclRefExpr that references a parameter.  We'll
988      //    add proper support for checking the attribute later.
989      if (HasVAListArg)
990        if (isa<ParmVarDecl>(VD))
991          return true;
992    }
993
994    return false;
995  }
996
997  case Stmt::CallExprClass: {
998    const CallExpr *CE = cast<CallExpr>(E);
999    if (const ImplicitCastExpr *ICE
1000          = dyn_cast<ImplicitCastExpr>(CE->getCallee())) {
1001      if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) {
1002        if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) {
1003          if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) {
1004            unsigned ArgIndex = FA->getFormatIdx();
1005            const Expr *Arg = CE->getArg(ArgIndex - 1);
1006
1007            return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg,
1008                                          format_idx, firstDataArg, isPrintf);
1009          }
1010        }
1011      }
1012    }
1013
1014    return false;
1015  }
1016  case Stmt::ObjCStringLiteralClass:
1017  case Stmt::StringLiteralClass: {
1018    const StringLiteral *StrE = NULL;
1019
1020    if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
1021      StrE = ObjCFExpr->getString();
1022    else
1023      StrE = cast<StringLiteral>(E);
1024
1025    if (StrE) {
1026      CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx,
1027                        firstDataArg, isPrintf);
1028      return true;
1029    }
1030
1031    return false;
1032  }
1033
1034  default:
1035    return false;
1036  }
1037}
1038
1039void
1040Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
1041                            const Expr * const *ExprArgs,
1042                            SourceLocation CallSiteLoc) {
1043  for (NonNullAttr::args_iterator i = NonNull->args_begin(),
1044                                  e = NonNull->args_end();
1045       i != e; ++i) {
1046    const Expr *ArgExpr = ExprArgs[*i];
1047    if (ArgExpr->isNullPointerConstant(Context,
1048                                       Expr::NPC_ValueDependentIsNotNull))
1049      Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange();
1050  }
1051}
1052
1053/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar
1054/// functions) for correct use of format strings.
1055void
1056Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg,
1057                                unsigned format_idx, unsigned firstDataArg,
1058                                bool isPrintf) {
1059
1060  const Expr *Fn = TheCall->getCallee();
1061
1062  // The way the format attribute works in GCC, the implicit this argument
1063  // of member functions is counted. However, it doesn't appear in our own
1064  // lists, so decrement format_idx in that case.
1065  if (isa<CXXMemberCallExpr>(TheCall)) {
1066    const CXXMethodDecl *method_decl =
1067      dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl());
1068    if (method_decl && method_decl->isInstance()) {
1069      // Catch a format attribute mistakenly referring to the object argument.
1070      if (format_idx == 0)
1071        return;
1072      --format_idx;
1073      if(firstDataArg != 0)
1074        --firstDataArg;
1075    }
1076  }
1077
1078  // CHECK: printf/scanf-like function is called with no format string.
1079  if (format_idx >= TheCall->getNumArgs()) {
1080    Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string)
1081      << Fn->getSourceRange();
1082    return;
1083  }
1084
1085  const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts();
1086
1087  // CHECK: format string is not a string literal.
1088  //
1089  // Dynamically generated format strings are difficult to
1090  // automatically vet at compile time.  Requiring that format strings
1091  // are string literals: (1) permits the checking of format strings by
1092  // the compiler and thereby (2) can practically remove the source of
1093  // many format string exploits.
1094
1095  // Format string can be either ObjC string (e.g. @"%d") or
1096  // C string (e.g. "%d")
1097  // ObjC string uses the same format specifiers as C string, so we can use
1098  // the same format string checking logic for both ObjC and C strings.
1099  if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx,
1100                             firstDataArg, isPrintf))
1101    return;  // Literal format string found, check done!
1102
1103  // If there are no arguments specified, warn with -Wformat-security, otherwise
1104  // warn only with -Wformat-nonliteral.
1105  if (TheCall->getNumArgs() == format_idx+1)
1106    Diag(TheCall->getArg(format_idx)->getLocStart(),
1107         diag::warn_format_nonliteral_noargs)
1108      << OrigFormatExpr->getSourceRange();
1109  else
1110    Diag(TheCall->getArg(format_idx)->getLocStart(),
1111         diag::warn_format_nonliteral)
1112           << OrigFormatExpr->getSourceRange();
1113}
1114
1115namespace {
1116class CheckFormatHandler : public analyze_format_string::FormatStringHandler {
1117protected:
1118  Sema &S;
1119  const StringLiteral *FExpr;
1120  const Expr *OrigFormatExpr;
1121  const unsigned FirstDataArg;
1122  const unsigned NumDataArgs;
1123  const bool IsObjCLiteral;
1124  const char *Beg; // Start of format string.
1125  const bool HasVAListArg;
1126  const CallExpr *TheCall;
1127  unsigned FormatIdx;
1128  llvm::BitVector CoveredArgs;
1129  bool usesPositionalArgs;
1130  bool atFirstArg;
1131public:
1132  CheckFormatHandler(Sema &s, const StringLiteral *fexpr,
1133                     const Expr *origFormatExpr, unsigned firstDataArg,
1134                     unsigned numDataArgs, bool isObjCLiteral,
1135                     const char *beg, bool hasVAListArg,
1136                     const CallExpr *theCall, unsigned formatIdx)
1137    : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr),
1138      FirstDataArg(firstDataArg),
1139      NumDataArgs(numDataArgs),
1140      IsObjCLiteral(isObjCLiteral), Beg(beg),
1141      HasVAListArg(hasVAListArg),
1142      TheCall(theCall), FormatIdx(formatIdx),
1143      usesPositionalArgs(false), atFirstArg(true) {
1144        CoveredArgs.resize(numDataArgs);
1145        CoveredArgs.reset();
1146      }
1147
1148  void DoneProcessing();
1149
1150  void HandleIncompleteSpecifier(const char *startSpecifier,
1151                                 unsigned specifierLen);
1152
1153  virtual void HandleInvalidPosition(const char *startSpecifier,
1154                                     unsigned specifierLen,
1155                                     analyze_format_string::PositionContext p);
1156
1157  virtual void HandleZeroPosition(const char *startPos, unsigned posLen);
1158
1159  void HandleNullChar(const char *nullCharacter);
1160
1161protected:
1162  bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc,
1163                                        const char *startSpec,
1164                                        unsigned specifierLen,
1165                                        const char *csStart, unsigned csLen);
1166
1167  SourceRange getFormatStringRange();
1168  CharSourceRange getSpecifierRange(const char *startSpecifier,
1169                                    unsigned specifierLen);
1170  SourceLocation getLocationOfByte(const char *x);
1171
1172  const Expr *getDataArg(unsigned i) const;
1173
1174  bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS,
1175                    const analyze_format_string::ConversionSpecifier &CS,
1176                    const char *startSpecifier, unsigned specifierLen,
1177                    unsigned argIndex);
1178};
1179}
1180
1181SourceRange CheckFormatHandler::getFormatStringRange() {
1182  return OrigFormatExpr->getSourceRange();
1183}
1184
1185CharSourceRange CheckFormatHandler::
1186getSpecifierRange(const char *startSpecifier, unsigned specifierLen) {
1187  SourceLocation Start = getLocationOfByte(startSpecifier);
1188  SourceLocation End   = getLocationOfByte(startSpecifier + specifierLen - 1);
1189
1190  // Advance the end SourceLocation by one due to half-open ranges.
1191  End = End.getFileLocWithOffset(1);
1192
1193  return CharSourceRange::getCharRange(Start, End);
1194}
1195
1196SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) {
1197  return S.getLocationOfStringLiteralByte(FExpr, x - Beg);
1198}
1199
1200void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier,
1201                                                   unsigned specifierLen){
1202  SourceLocation Loc = getLocationOfByte(startSpecifier);
1203  S.Diag(Loc, diag::warn_printf_incomplete_specifier)
1204    << getSpecifierRange(startSpecifier, specifierLen);
1205}
1206
1207void
1208CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen,
1209                                     analyze_format_string::PositionContext p) {
1210  SourceLocation Loc = getLocationOfByte(startPos);
1211  S.Diag(Loc, diag::warn_format_invalid_positional_specifier)
1212    << (unsigned) p << getSpecifierRange(startPos, posLen);
1213}
1214
1215void CheckFormatHandler::HandleZeroPosition(const char *startPos,
1216                                            unsigned posLen) {
1217  SourceLocation Loc = getLocationOfByte(startPos);
1218  S.Diag(Loc, diag::warn_format_zero_positional_specifier)
1219    << getSpecifierRange(startPos, posLen);
1220}
1221
1222void CheckFormatHandler::HandleNullChar(const char *nullCharacter) {
1223  if (!IsObjCLiteral) {
1224    // The presence of a null character is likely an error.
1225    S.Diag(getLocationOfByte(nullCharacter),
1226           diag::warn_printf_format_string_contains_null_char)
1227      << getFormatStringRange();
1228  }
1229}
1230
1231const Expr *CheckFormatHandler::getDataArg(unsigned i) const {
1232  return TheCall->getArg(FirstDataArg + i);
1233}
1234
1235void CheckFormatHandler::DoneProcessing() {
1236    // Does the number of data arguments exceed the number of
1237    // format conversions in the format string?
1238  if (!HasVAListArg) {
1239      // Find any arguments that weren't covered.
1240    CoveredArgs.flip();
1241    signed notCoveredArg = CoveredArgs.find_first();
1242    if (notCoveredArg >= 0) {
1243      assert((unsigned)notCoveredArg < NumDataArgs);
1244      S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(),
1245             diag::warn_printf_data_arg_not_used)
1246      << getFormatStringRange();
1247    }
1248  }
1249}
1250
1251bool
1252CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex,
1253                                                     SourceLocation Loc,
1254                                                     const char *startSpec,
1255                                                     unsigned specifierLen,
1256                                                     const char *csStart,
1257                                                     unsigned csLen) {
1258
1259  bool keepGoing = true;
1260  if (argIndex < NumDataArgs) {
1261    // Consider the argument coverered, even though the specifier doesn't
1262    // make sense.
1263    CoveredArgs.set(argIndex);
1264  }
1265  else {
1266    // If argIndex exceeds the number of data arguments we
1267    // don't issue a warning because that is just a cascade of warnings (and
1268    // they may have intended '%%' anyway). We don't want to continue processing
1269    // the format string after this point, however, as we will like just get
1270    // gibberish when trying to match arguments.
1271    keepGoing = false;
1272  }
1273
1274  S.Diag(Loc, diag::warn_format_invalid_conversion)
1275    << llvm::StringRef(csStart, csLen)
1276    << getSpecifierRange(startSpec, specifierLen);
1277
1278  return keepGoing;
1279}
1280
1281bool
1282CheckFormatHandler::CheckNumArgs(
1283  const analyze_format_string::FormatSpecifier &FS,
1284  const analyze_format_string::ConversionSpecifier &CS,
1285  const char *startSpecifier, unsigned specifierLen, unsigned argIndex) {
1286
1287  if (argIndex >= NumDataArgs) {
1288    if (FS.usesPositionalArg())  {
1289      S.Diag(getLocationOfByte(CS.getStart()),
1290             diag::warn_printf_positional_arg_exceeds_data_args)
1291      << (argIndex+1) << NumDataArgs
1292      << getSpecifierRange(startSpecifier, specifierLen);
1293    }
1294    else {
1295      S.Diag(getLocationOfByte(CS.getStart()),
1296             diag::warn_printf_insufficient_data_args)
1297      << getSpecifierRange(startSpecifier, specifierLen);
1298    }
1299
1300    return false;
1301  }
1302  return true;
1303}
1304
1305//===--- CHECK: Printf format string checking ------------------------------===//
1306
1307namespace {
1308class CheckPrintfHandler : public CheckFormatHandler {
1309public:
1310  CheckPrintfHandler(Sema &s, const StringLiteral *fexpr,
1311                     const Expr *origFormatExpr, unsigned firstDataArg,
1312                     unsigned numDataArgs, bool isObjCLiteral,
1313                     const char *beg, bool hasVAListArg,
1314                     const CallExpr *theCall, unsigned formatIdx)
1315  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1316                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1317                       theCall, formatIdx) {}
1318
1319
1320  bool HandleInvalidPrintfConversionSpecifier(
1321                                      const analyze_printf::PrintfSpecifier &FS,
1322                                      const char *startSpecifier,
1323                                      unsigned specifierLen);
1324
1325  bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS,
1326                             const char *startSpecifier,
1327                             unsigned specifierLen);
1328
1329  bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k,
1330                    const char *startSpecifier, unsigned specifierLen);
1331  void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS,
1332                           const analyze_printf::OptionalAmount &Amt,
1333                           unsigned type,
1334                           const char *startSpecifier, unsigned specifierLen);
1335  void HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1336                  const analyze_printf::OptionalFlag &flag,
1337                  const char *startSpecifier, unsigned specifierLen);
1338  void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS,
1339                         const analyze_printf::OptionalFlag &ignoredFlag,
1340                         const analyze_printf::OptionalFlag &flag,
1341                         const char *startSpecifier, unsigned specifierLen);
1342};
1343}
1344
1345bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier(
1346                                      const analyze_printf::PrintfSpecifier &FS,
1347                                      const char *startSpecifier,
1348                                      unsigned specifierLen) {
1349  const analyze_printf::PrintfConversionSpecifier &CS =
1350    FS.getConversionSpecifier();
1351
1352  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1353                                          getLocationOfByte(CS.getStart()),
1354                                          startSpecifier, specifierLen,
1355                                          CS.getStart(), CS.getLength());
1356}
1357
1358bool CheckPrintfHandler::HandleAmount(
1359                               const analyze_format_string::OptionalAmount &Amt,
1360                               unsigned k, const char *startSpecifier,
1361                               unsigned specifierLen) {
1362
1363  if (Amt.hasDataArgument()) {
1364    if (!HasVAListArg) {
1365      unsigned argIndex = Amt.getArgIndex();
1366      if (argIndex >= NumDataArgs) {
1367        S.Diag(getLocationOfByte(Amt.getStart()),
1368               diag::warn_printf_asterisk_missing_arg)
1369          << k << getSpecifierRange(startSpecifier, specifierLen);
1370        // Don't do any more checking.  We will just emit
1371        // spurious errors.
1372        return false;
1373      }
1374
1375      // Type check the data argument.  It should be an 'int'.
1376      // Although not in conformance with C99, we also allow the argument to be
1377      // an 'unsigned int' as that is a reasonably safe case.  GCC also
1378      // doesn't emit a warning for that case.
1379      CoveredArgs.set(argIndex);
1380      const Expr *Arg = getDataArg(argIndex);
1381      QualType T = Arg->getType();
1382
1383      const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context);
1384      assert(ATR.isValid());
1385
1386      if (!ATR.matchesType(S.Context, T)) {
1387        S.Diag(getLocationOfByte(Amt.getStart()),
1388               diag::warn_printf_asterisk_wrong_type)
1389          << k
1390          << ATR.getRepresentativeType(S.Context) << T
1391          << getSpecifierRange(startSpecifier, specifierLen)
1392          << Arg->getSourceRange();
1393        // Don't do any more checking.  We will just emit
1394        // spurious errors.
1395        return false;
1396      }
1397    }
1398  }
1399  return true;
1400}
1401
1402void CheckPrintfHandler::HandleInvalidAmount(
1403                                      const analyze_printf::PrintfSpecifier &FS,
1404                                      const analyze_printf::OptionalAmount &Amt,
1405                                      unsigned type,
1406                                      const char *startSpecifier,
1407                                      unsigned specifierLen) {
1408  const analyze_printf::PrintfConversionSpecifier &CS =
1409    FS.getConversionSpecifier();
1410  switch (Amt.getHowSpecified()) {
1411  case analyze_printf::OptionalAmount::Constant:
1412    S.Diag(getLocationOfByte(Amt.getStart()),
1413        diag::warn_printf_nonsensical_optional_amount)
1414      << type
1415      << CS.toString()
1416      << getSpecifierRange(startSpecifier, specifierLen)
1417      << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(),
1418          Amt.getConstantLength()));
1419    break;
1420
1421  default:
1422    S.Diag(getLocationOfByte(Amt.getStart()),
1423        diag::warn_printf_nonsensical_optional_amount)
1424      << type
1425      << CS.toString()
1426      << getSpecifierRange(startSpecifier, specifierLen);
1427    break;
1428  }
1429}
1430
1431void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1432                                    const analyze_printf::OptionalFlag &flag,
1433                                    const char *startSpecifier,
1434                                    unsigned specifierLen) {
1435  // Warn about pointless flag with a fixit removal.
1436  const analyze_printf::PrintfConversionSpecifier &CS =
1437    FS.getConversionSpecifier();
1438  S.Diag(getLocationOfByte(flag.getPosition()),
1439      diag::warn_printf_nonsensical_flag)
1440    << flag.toString() << CS.toString()
1441    << getSpecifierRange(startSpecifier, specifierLen)
1442    << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1));
1443}
1444
1445void CheckPrintfHandler::HandleIgnoredFlag(
1446                                const analyze_printf::PrintfSpecifier &FS,
1447                                const analyze_printf::OptionalFlag &ignoredFlag,
1448                                const analyze_printf::OptionalFlag &flag,
1449                                const char *startSpecifier,
1450                                unsigned specifierLen) {
1451  // Warn about ignored flag with a fixit removal.
1452  S.Diag(getLocationOfByte(ignoredFlag.getPosition()),
1453      diag::warn_printf_ignored_flag)
1454    << ignoredFlag.toString() << flag.toString()
1455    << getSpecifierRange(startSpecifier, specifierLen)
1456    << FixItHint::CreateRemoval(getSpecifierRange(
1457        ignoredFlag.getPosition(), 1));
1458}
1459
1460bool
1461CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
1462                                            &FS,
1463                                          const char *startSpecifier,
1464                                          unsigned specifierLen) {
1465
1466  using namespace analyze_format_string;
1467  using namespace analyze_printf;
1468  const PrintfConversionSpecifier &CS = FS.getConversionSpecifier();
1469
1470  if (FS.consumesDataArgument()) {
1471    if (atFirstArg) {
1472        atFirstArg = false;
1473        usesPositionalArgs = FS.usesPositionalArg();
1474    }
1475    else if (usesPositionalArgs != FS.usesPositionalArg()) {
1476      // Cannot mix-and-match positional and non-positional arguments.
1477      S.Diag(getLocationOfByte(CS.getStart()),
1478             diag::warn_format_mix_positional_nonpositional_args)
1479        << getSpecifierRange(startSpecifier, specifierLen);
1480      return false;
1481    }
1482  }
1483
1484  // First check if the field width, precision, and conversion specifier
1485  // have matching data arguments.
1486  if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0,
1487                    startSpecifier, specifierLen)) {
1488    return false;
1489  }
1490
1491  if (!HandleAmount(FS.getPrecision(), /* precision */ 1,
1492                    startSpecifier, specifierLen)) {
1493    return false;
1494  }
1495
1496  if (!CS.consumesDataArgument()) {
1497    // FIXME: Technically specifying a precision or field width here
1498    // makes no sense.  Worth issuing a warning at some point.
1499    return true;
1500  }
1501
1502  // Consume the argument.
1503  unsigned argIndex = FS.getArgIndex();
1504  if (argIndex < NumDataArgs) {
1505    // The check to see if the argIndex is valid will come later.
1506    // We set the bit here because we may exit early from this
1507    // function if we encounter some other error.
1508    CoveredArgs.set(argIndex);
1509  }
1510
1511  // Check for using an Objective-C specific conversion specifier
1512  // in a non-ObjC literal.
1513  if (!IsObjCLiteral && CS.isObjCArg()) {
1514    return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier,
1515                                                  specifierLen);
1516  }
1517
1518  // Check for invalid use of field width
1519  if (!FS.hasValidFieldWidth()) {
1520    HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0,
1521        startSpecifier, specifierLen);
1522  }
1523
1524  // Check for invalid use of precision
1525  if (!FS.hasValidPrecision()) {
1526    HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1,
1527        startSpecifier, specifierLen);
1528  }
1529
1530  // Check each flag does not conflict with any other component.
1531  if (!FS.hasValidThousandsGroupingPrefix())
1532    HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen);
1533  if (!FS.hasValidLeadingZeros())
1534    HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen);
1535  if (!FS.hasValidPlusPrefix())
1536    HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen);
1537  if (!FS.hasValidSpacePrefix())
1538    HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen);
1539  if (!FS.hasValidAlternativeForm())
1540    HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen);
1541  if (!FS.hasValidLeftJustified())
1542    HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen);
1543
1544  // Check that flags are not ignored by another flag
1545  if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+'
1546    HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(),
1547        startSpecifier, specifierLen);
1548  if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-'
1549    HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(),
1550            startSpecifier, specifierLen);
1551
1552  // Check the length modifier is valid with the given conversion specifier.
1553  const LengthModifier &LM = FS.getLengthModifier();
1554  if (!FS.hasValidLengthModifier())
1555    S.Diag(getLocationOfByte(LM.getStart()),
1556        diag::warn_format_nonsensical_length)
1557      << LM.toString() << CS.toString()
1558      << getSpecifierRange(startSpecifier, specifierLen)
1559      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1560          LM.getLength()));
1561
1562  // Are we using '%n'?
1563  if (CS.getKind() == ConversionSpecifier::nArg) {
1564    // Issue a warning about this being a possible security issue.
1565    S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back)
1566      << getSpecifierRange(startSpecifier, specifierLen);
1567    // Continue checking the other format specifiers.
1568    return true;
1569  }
1570
1571  // The remaining checks depend on the data arguments.
1572  if (HasVAListArg)
1573    return true;
1574
1575  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
1576    return false;
1577
1578  // Now type check the data expression that matches the
1579  // format specifier.
1580  const Expr *Ex = getDataArg(argIndex);
1581  const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context);
1582  if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) {
1583    // Check if we didn't match because of an implicit cast from a 'char'
1584    // or 'short' to an 'int'.  This is done because printf is a varargs
1585    // function.
1586    if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex))
1587      if (ICE->getType() == S.Context.IntTy) {
1588        // All further checking is done on the subexpression.
1589        Ex = ICE->getSubExpr();
1590        if (ATR.matchesType(S.Context, Ex->getType()))
1591          return true;
1592      }
1593
1594    // We may be able to offer a FixItHint if it is a supported type.
1595    PrintfSpecifier fixedFS = FS;
1596    bool success = fixedFS.fixType(Ex->getType());
1597
1598    if (success) {
1599      // Get the fix string from the fixed format specifier
1600      llvm::SmallString<128> buf;
1601      llvm::raw_svector_ostream os(buf);
1602      fixedFS.toString(os);
1603
1604      // FIXME: getRepresentativeType() perhaps should return a string
1605      // instead of a QualType to better handle when the representative
1606      // type is 'wint_t' (which is defined in the system headers).
1607      S.Diag(getLocationOfByte(CS.getStart()),
1608          diag::warn_printf_conversion_argument_type_mismatch)
1609        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1610        << getSpecifierRange(startSpecifier, specifierLen)
1611        << Ex->getSourceRange()
1612        << FixItHint::CreateReplacement(
1613            getSpecifierRange(startSpecifier, specifierLen),
1614            os.str());
1615    }
1616    else {
1617      S.Diag(getLocationOfByte(CS.getStart()),
1618             diag::warn_printf_conversion_argument_type_mismatch)
1619        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1620        << getSpecifierRange(startSpecifier, specifierLen)
1621        << Ex->getSourceRange();
1622    }
1623  }
1624
1625  return true;
1626}
1627
1628//===--- CHECK: Scanf format string checking ------------------------------===//
1629
1630namespace {
1631class CheckScanfHandler : public CheckFormatHandler {
1632public:
1633  CheckScanfHandler(Sema &s, const StringLiteral *fexpr,
1634                    const Expr *origFormatExpr, unsigned firstDataArg,
1635                    unsigned numDataArgs, bool isObjCLiteral,
1636                    const char *beg, bool hasVAListArg,
1637                    const CallExpr *theCall, unsigned formatIdx)
1638  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1639                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1640                       theCall, formatIdx) {}
1641
1642  bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS,
1643                            const char *startSpecifier,
1644                            unsigned specifierLen);
1645
1646  bool HandleInvalidScanfConversionSpecifier(
1647          const analyze_scanf::ScanfSpecifier &FS,
1648          const char *startSpecifier,
1649          unsigned specifierLen);
1650
1651  void HandleIncompleteScanList(const char *start, const char *end);
1652};
1653}
1654
1655void CheckScanfHandler::HandleIncompleteScanList(const char *start,
1656                                                 const char *end) {
1657  S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete)
1658    << getSpecifierRange(start, end - start);
1659}
1660
1661bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier(
1662                                        const analyze_scanf::ScanfSpecifier &FS,
1663                                        const char *startSpecifier,
1664                                        unsigned specifierLen) {
1665
1666  const analyze_scanf::ScanfConversionSpecifier &CS =
1667    FS.getConversionSpecifier();
1668
1669  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1670                                          getLocationOfByte(CS.getStart()),
1671                                          startSpecifier, specifierLen,
1672                                          CS.getStart(), CS.getLength());
1673}
1674
1675bool CheckScanfHandler::HandleScanfSpecifier(
1676                                       const analyze_scanf::ScanfSpecifier &FS,
1677                                       const char *startSpecifier,
1678                                       unsigned specifierLen) {
1679
1680  using namespace analyze_scanf;
1681  using namespace analyze_format_string;
1682
1683  const ScanfConversionSpecifier &CS = FS.getConversionSpecifier();
1684
1685  // Handle case where '%' and '*' don't consume an argument.  These shouldn't
1686  // be used to decide if we are using positional arguments consistently.
1687  if (FS.consumesDataArgument()) {
1688    if (atFirstArg) {
1689      atFirstArg = false;
1690      usesPositionalArgs = FS.usesPositionalArg();
1691    }
1692    else if (usesPositionalArgs != FS.usesPositionalArg()) {
1693      // Cannot mix-and-match positional and non-positional arguments.
1694      S.Diag(getLocationOfByte(CS.getStart()),
1695             diag::warn_format_mix_positional_nonpositional_args)
1696        << getSpecifierRange(startSpecifier, specifierLen);
1697      return false;
1698    }
1699  }
1700
1701  // Check if the field with is non-zero.
1702  const OptionalAmount &Amt = FS.getFieldWidth();
1703  if (Amt.getHowSpecified() == OptionalAmount::Constant) {
1704    if (Amt.getConstantAmount() == 0) {
1705      const CharSourceRange &R = getSpecifierRange(Amt.getStart(),
1706                                                   Amt.getConstantLength());
1707      S.Diag(getLocationOfByte(Amt.getStart()),
1708             diag::warn_scanf_nonzero_width)
1709        << R << FixItHint::CreateRemoval(R);
1710    }
1711  }
1712
1713  if (!FS.consumesDataArgument()) {
1714    // FIXME: Technically specifying a precision or field width here
1715    // makes no sense.  Worth issuing a warning at some point.
1716    return true;
1717  }
1718
1719  // Consume the argument.
1720  unsigned argIndex = FS.getArgIndex();
1721  if (argIndex < NumDataArgs) {
1722      // The check to see if the argIndex is valid will come later.
1723      // We set the bit here because we may exit early from this
1724      // function if we encounter some other error.
1725    CoveredArgs.set(argIndex);
1726  }
1727
1728  // Check the length modifier is valid with the given conversion specifier.
1729  const LengthModifier &LM = FS.getLengthModifier();
1730  if (!FS.hasValidLengthModifier()) {
1731    S.Diag(getLocationOfByte(LM.getStart()),
1732           diag::warn_format_nonsensical_length)
1733      << LM.toString() << CS.toString()
1734      << getSpecifierRange(startSpecifier, specifierLen)
1735      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1736                                                    LM.getLength()));
1737  }
1738
1739  // The remaining checks depend on the data arguments.
1740  if (HasVAListArg)
1741    return true;
1742
1743  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
1744    return false;
1745
1746  // FIXME: Check that the argument type matches the format specifier.
1747
1748  return true;
1749}
1750
1751void Sema::CheckFormatString(const StringLiteral *FExpr,
1752                             const Expr *OrigFormatExpr,
1753                             const CallExpr *TheCall, bool HasVAListArg,
1754                             unsigned format_idx, unsigned firstDataArg,
1755                             bool isPrintf) {
1756
1757  // CHECK: is the format string a wide literal?
1758  if (FExpr->isWide()) {
1759    Diag(FExpr->getLocStart(),
1760         diag::warn_format_string_is_wide_literal)
1761    << OrigFormatExpr->getSourceRange();
1762    return;
1763  }
1764
1765  // Str - The format string.  NOTE: this is NOT null-terminated!
1766  llvm::StringRef StrRef = FExpr->getString();
1767  const char *Str = StrRef.data();
1768  unsigned StrLen = StrRef.size();
1769
1770  // CHECK: empty format string?
1771  if (StrLen == 0) {
1772    Diag(FExpr->getLocStart(), diag::warn_empty_format_string)
1773    << OrigFormatExpr->getSourceRange();
1774    return;
1775  }
1776
1777  if (isPrintf) {
1778    CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1779                         TheCall->getNumArgs() - firstDataArg,
1780                         isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1781                         HasVAListArg, TheCall, format_idx);
1782
1783    if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen))
1784      H.DoneProcessing();
1785  }
1786  else {
1787    CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1788                        TheCall->getNumArgs() - firstDataArg,
1789                        isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1790                        HasVAListArg, TheCall, format_idx);
1791
1792    if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen))
1793      H.DoneProcessing();
1794  }
1795}
1796
1797//===--- CHECK: Return Address of Stack Variable --------------------------===//
1798
1799static Expr *EvalVal(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars);
1800static Expr *EvalAddr(Expr* E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars);
1801
1802/// CheckReturnStackAddr - Check if a return statement returns the address
1803///   of a stack variable.
1804void
1805Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
1806                           SourceLocation ReturnLoc) {
1807
1808  Expr *stackE = 0;
1809  llvm::SmallVector<DeclRefExpr *, 8> refVars;
1810
1811  // Perform checking for returned stack addresses, local blocks,
1812  // label addresses or references to temporaries.
1813  if (lhsType->isPointerType() || lhsType->isBlockPointerType()) {
1814    stackE = EvalAddr(RetValExp, refVars);
1815  } else if (lhsType->isReferenceType()) {
1816    stackE = EvalVal(RetValExp, refVars);
1817  }
1818
1819  if (stackE == 0)
1820    return; // Nothing suspicious was found.
1821
1822  SourceLocation diagLoc;
1823  SourceRange diagRange;
1824  if (refVars.empty()) {
1825    diagLoc = stackE->getLocStart();
1826    diagRange = stackE->getSourceRange();
1827  } else {
1828    // We followed through a reference variable. 'stackE' contains the
1829    // problematic expression but we will warn at the return statement pointing
1830    // at the reference variable. We will later display the "trail" of
1831    // reference variables using notes.
1832    diagLoc = refVars[0]->getLocStart();
1833    diagRange = refVars[0]->getSourceRange();
1834  }
1835
1836  if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var.
1837    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref
1838                                             : diag::warn_ret_stack_addr)
1839     << DR->getDecl()->getDeclName() << diagRange;
1840  } else if (isa<BlockExpr>(stackE)) { // local block.
1841    Diag(diagLoc, diag::err_ret_local_block) << diagRange;
1842  } else if (isa<AddrLabelExpr>(stackE)) { // address of label.
1843    Diag(diagLoc, diag::warn_ret_addr_label) << diagRange;
1844  } else { // local temporary.
1845    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref
1846                                             : diag::warn_ret_local_temp_addr)
1847     << diagRange;
1848  }
1849
1850  // Display the "trail" of reference variables that we followed until we
1851  // found the problematic expression using notes.
1852  for (unsigned i = 0, e = refVars.size(); i != e; ++i) {
1853    VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl());
1854    // If this var binds to another reference var, show the range of the next
1855    // var, otherwise the var binds to the problematic expression, in which case
1856    // show the range of the expression.
1857    SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange()
1858                                  : stackE->getSourceRange();
1859    Diag(VD->getLocation(), diag::note_ref_var_local_bind)
1860      << VD->getDeclName() << range;
1861  }
1862}
1863
1864/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
1865///  check if the expression in a return statement evaluates to an address
1866///  to a location on the stack, a local block, an address of a label, or a
1867///  reference to local temporary. The recursion is used to traverse the
1868///  AST of the return expression, with recursion backtracking when we
1869///  encounter a subexpression that (1) clearly does not lead to one of the
1870///  above problematic expressions (2) is something we cannot determine leads to
1871///  a problematic expression based on such local checking.
1872///
1873///  Both EvalAddr and EvalVal follow through reference variables to evaluate
1874///  the expression that they point to. Such variables are added to the
1875///  'refVars' vector so that we know what the reference variable "trail" was.
1876///
1877///  EvalAddr processes expressions that are pointers that are used as
1878///  references (and not L-values).  EvalVal handles all other values.
1879///  At the base case of the recursion is a check for the above problematic
1880///  expressions.
1881///
1882///  This implementation handles:
1883///
1884///   * pointer-to-pointer casts
1885///   * implicit conversions from array references to pointers
1886///   * taking the address of fields
1887///   * arbitrary interplay between "&" and "*" operators
1888///   * pointer arithmetic from an address of a stack variable
1889///   * taking the address of an array element where the array is on the stack
1890static Expr *EvalAddr(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars) {
1891  if (E->isTypeDependent())
1892      return NULL;
1893
1894  // We should only be called for evaluating pointer expressions.
1895  assert((E->getType()->isAnyPointerType() ||
1896          E->getType()->isBlockPointerType() ||
1897          E->getType()->isObjCQualifiedIdType()) &&
1898         "EvalAddr only works on pointers");
1899
1900  // Our "symbolic interpreter" is just a dispatch off the currently
1901  // viewed AST node.  We then recursively traverse the AST by calling
1902  // EvalAddr and EvalVal appropriately.
1903  switch (E->getStmtClass()) {
1904  case Stmt::ParenExprClass:
1905    // Ignore parentheses.
1906    return EvalAddr(cast<ParenExpr>(E)->getSubExpr(), refVars);
1907
1908  case Stmt::DeclRefExprClass: {
1909    DeclRefExpr *DR = cast<DeclRefExpr>(E);
1910
1911    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
1912      // If this is a reference variable, follow through to the expression that
1913      // it points to.
1914      if (V->hasLocalStorage() &&
1915          V->getType()->isReferenceType() && V->hasInit()) {
1916        // Add the reference variable to the "trail".
1917        refVars.push_back(DR);
1918        return EvalAddr(V->getInit(), refVars);
1919      }
1920
1921    return NULL;
1922  }
1923
1924  case Stmt::UnaryOperatorClass: {
1925    // The only unary operator that make sense to handle here
1926    // is AddrOf.  All others don't make sense as pointers.
1927    UnaryOperator *U = cast<UnaryOperator>(E);
1928
1929    if (U->getOpcode() == UO_AddrOf)
1930      return EvalVal(U->getSubExpr(), refVars);
1931    else
1932      return NULL;
1933  }
1934
1935  case Stmt::BinaryOperatorClass: {
1936    // Handle pointer arithmetic.  All other binary operators are not valid
1937    // in this context.
1938    BinaryOperator *B = cast<BinaryOperator>(E);
1939    BinaryOperatorKind op = B->getOpcode();
1940
1941    if (op != BO_Add && op != BO_Sub)
1942      return NULL;
1943
1944    Expr *Base = B->getLHS();
1945
1946    // Determine which argument is the real pointer base.  It could be
1947    // the RHS argument instead of the LHS.
1948    if (!Base->getType()->isPointerType()) Base = B->getRHS();
1949
1950    assert (Base->getType()->isPointerType());
1951    return EvalAddr(Base, refVars);
1952  }
1953
1954  // For conditional operators we need to see if either the LHS or RHS are
1955  // valid DeclRefExpr*s.  If one of them is valid, we return it.
1956  case Stmt::ConditionalOperatorClass: {
1957    ConditionalOperator *C = cast<ConditionalOperator>(E);
1958
1959    // Handle the GNU extension for missing LHS.
1960    if (Expr *lhsExpr = C->getLHS()) {
1961    // In C++, we can have a throw-expression, which has 'void' type.
1962      if (!lhsExpr->getType()->isVoidType())
1963        if (Expr* LHS = EvalAddr(lhsExpr, refVars))
1964          return LHS;
1965    }
1966
1967    // In C++, we can have a throw-expression, which has 'void' type.
1968    if (C->getRHS()->getType()->isVoidType())
1969      return NULL;
1970
1971    return EvalAddr(C->getRHS(), refVars);
1972  }
1973
1974  case Stmt::BlockExprClass:
1975    if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures())
1976      return E; // local block.
1977    return NULL;
1978
1979  case Stmt::AddrLabelExprClass:
1980    return E; // address of label.
1981
1982  // For casts, we need to handle conversions from arrays to
1983  // pointer values, and pointer-to-pointer conversions.
1984  case Stmt::ImplicitCastExprClass:
1985  case Stmt::CStyleCastExprClass:
1986  case Stmt::CXXFunctionalCastExprClass: {
1987    Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
1988    QualType T = SubExpr->getType();
1989
1990    if (SubExpr->getType()->isPointerType() ||
1991        SubExpr->getType()->isBlockPointerType() ||
1992        SubExpr->getType()->isObjCQualifiedIdType())
1993      return EvalAddr(SubExpr, refVars);
1994    else if (T->isArrayType())
1995      return EvalVal(SubExpr, refVars);
1996    else
1997      return 0;
1998  }
1999
2000  // C++ casts.  For dynamic casts, static casts, and const casts, we
2001  // are always converting from a pointer-to-pointer, so we just blow
2002  // through the cast.  In the case the dynamic cast doesn't fail (and
2003  // return NULL), we take the conservative route and report cases
2004  // where we return the address of a stack variable.  For Reinterpre
2005  // FIXME: The comment about is wrong; we're not always converting
2006  // from pointer to pointer. I'm guessing that this code should also
2007  // handle references to objects.
2008  case Stmt::CXXStaticCastExprClass:
2009  case Stmt::CXXDynamicCastExprClass:
2010  case Stmt::CXXConstCastExprClass:
2011  case Stmt::CXXReinterpretCastExprClass: {
2012      Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr();
2013      if (S->getType()->isPointerType() || S->getType()->isBlockPointerType())
2014        return EvalAddr(S, refVars);
2015      else
2016        return NULL;
2017  }
2018
2019  // Everything else: we simply don't reason about them.
2020  default:
2021    return NULL;
2022  }
2023}
2024
2025
2026///  EvalVal - This function is complements EvalAddr in the mutual recursion.
2027///   See the comments for EvalAddr for more details.
2028static Expr *EvalVal(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars) {
2029do {
2030  // We should only be called for evaluating non-pointer expressions, or
2031  // expressions with a pointer type that are not used as references but instead
2032  // are l-values (e.g., DeclRefExpr with a pointer type).
2033
2034  // Our "symbolic interpreter" is just a dispatch off the currently
2035  // viewed AST node.  We then recursively traverse the AST by calling
2036  // EvalAddr and EvalVal appropriately.
2037  switch (E->getStmtClass()) {
2038  case Stmt::ImplicitCastExprClass: {
2039    ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E);
2040    if (IE->getValueKind() == VK_LValue) {
2041      E = IE->getSubExpr();
2042      continue;
2043    }
2044    return NULL;
2045  }
2046
2047  case Stmt::DeclRefExprClass: {
2048    // When we hit a DeclRefExpr we are looking at code that refers to a
2049    // variable's name. If it's not a reference variable we check if it has
2050    // local storage within the function, and if so, return the expression.
2051    DeclRefExpr *DR = cast<DeclRefExpr>(E);
2052
2053    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
2054      if (V->hasLocalStorage()) {
2055        if (!V->getType()->isReferenceType())
2056          return DR;
2057
2058        // Reference variable, follow through to the expression that
2059        // it points to.
2060        if (V->hasInit()) {
2061          // Add the reference variable to the "trail".
2062          refVars.push_back(DR);
2063          return EvalVal(V->getInit(), refVars);
2064        }
2065      }
2066
2067    return NULL;
2068  }
2069
2070  case Stmt::ParenExprClass: {
2071    // Ignore parentheses.
2072    E = cast<ParenExpr>(E)->getSubExpr();
2073    continue;
2074  }
2075
2076  case Stmt::UnaryOperatorClass: {
2077    // The only unary operator that make sense to handle here
2078    // is Deref.  All others don't resolve to a "name."  This includes
2079    // handling all sorts of rvalues passed to a unary operator.
2080    UnaryOperator *U = cast<UnaryOperator>(E);
2081
2082    if (U->getOpcode() == UO_Deref)
2083      return EvalAddr(U->getSubExpr(), refVars);
2084
2085    return NULL;
2086  }
2087
2088  case Stmt::ArraySubscriptExprClass: {
2089    // Array subscripts are potential references to data on the stack.  We
2090    // retrieve the DeclRefExpr* for the array variable if it indeed
2091    // has local storage.
2092    return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars);
2093  }
2094
2095  case Stmt::ConditionalOperatorClass: {
2096    // For conditional operators we need to see if either the LHS or RHS are
2097    // non-NULL Expr's.  If one is non-NULL, we return it.
2098    ConditionalOperator *C = cast<ConditionalOperator>(E);
2099
2100    // Handle the GNU extension for missing LHS.
2101    if (Expr *lhsExpr = C->getLHS())
2102      if (Expr *LHS = EvalVal(lhsExpr, refVars))
2103        return LHS;
2104
2105    return EvalVal(C->getRHS(), refVars);
2106  }
2107
2108  // Accesses to members are potential references to data on the stack.
2109  case Stmt::MemberExprClass: {
2110    MemberExpr *M = cast<MemberExpr>(E);
2111
2112    // Check for indirect access.  We only want direct field accesses.
2113    if (M->isArrow())
2114      return NULL;
2115
2116    // Check whether the member type is itself a reference, in which case
2117    // we're not going to refer to the member, but to what the member refers to.
2118    if (M->getMemberDecl()->getType()->isReferenceType())
2119      return NULL;
2120
2121    return EvalVal(M->getBase(), refVars);
2122  }
2123
2124  default:
2125    // Check that we don't return or take the address of a reference to a
2126    // temporary. This is only useful in C++.
2127    if (!E->isTypeDependent() && E->isRValue())
2128      return E;
2129
2130    // Everything else: we simply don't reason about them.
2131    return NULL;
2132  }
2133} while (true);
2134}
2135
2136//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
2137
2138/// Check for comparisons of floating point operands using != and ==.
2139/// Issue a warning if these are no self-comparisons, as they are not likely
2140/// to do what the programmer intended.
2141void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) {
2142  bool EmitWarning = true;
2143
2144  Expr* LeftExprSansParen = lex->IgnoreParenImpCasts();
2145  Expr* RightExprSansParen = rex->IgnoreParenImpCasts();
2146
2147  // Special case: check for x == x (which is OK).
2148  // Do not emit warnings for such cases.
2149  if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
2150    if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
2151      if (DRL->getDecl() == DRR->getDecl())
2152        EmitWarning = false;
2153
2154
2155  // Special case: check for comparisons against literals that can be exactly
2156  //  represented by APFloat.  In such cases, do not emit a warning.  This
2157  //  is a heuristic: often comparison against such literals are used to
2158  //  detect if a value in a variable has not changed.  This clearly can
2159  //  lead to false negatives.
2160  if (EmitWarning) {
2161    if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
2162      if (FLL->isExact())
2163        EmitWarning = false;
2164    } else
2165      if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){
2166        if (FLR->isExact())
2167          EmitWarning = false;
2168    }
2169  }
2170
2171  // Check for comparisons with builtin types.
2172  if (EmitWarning)
2173    if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
2174      if (CL->isBuiltinCall(Context))
2175        EmitWarning = false;
2176
2177  if (EmitWarning)
2178    if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
2179      if (CR->isBuiltinCall(Context))
2180        EmitWarning = false;
2181
2182  // Emit the diagnostic.
2183  if (EmitWarning)
2184    Diag(loc, diag::warn_floatingpoint_eq)
2185      << lex->getSourceRange() << rex->getSourceRange();
2186}
2187
2188//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===//
2189//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===//
2190
2191namespace {
2192
2193/// Structure recording the 'active' range of an integer-valued
2194/// expression.
2195struct IntRange {
2196  /// The number of bits active in the int.
2197  unsigned Width;
2198
2199  /// True if the int is known not to have negative values.
2200  bool NonNegative;
2201
2202  IntRange(unsigned Width, bool NonNegative)
2203    : Width(Width), NonNegative(NonNegative)
2204  {}
2205
2206  /// Returns the range of the bool type.
2207  static IntRange forBoolType() {
2208    return IntRange(1, true);
2209  }
2210
2211  /// Returns the range of an opaque value of the given integral type.
2212  static IntRange forValueOfType(ASTContext &C, QualType T) {
2213    return forValueOfCanonicalType(C,
2214                          T->getCanonicalTypeInternal().getTypePtr());
2215  }
2216
2217  /// Returns the range of an opaque value of a canonical integral type.
2218  static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) {
2219    assert(T->isCanonicalUnqualified());
2220
2221    if (const VectorType *VT = dyn_cast<VectorType>(T))
2222      T = VT->getElementType().getTypePtr();
2223    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2224      T = CT->getElementType().getTypePtr();
2225
2226    // For enum types, use the known bit width of the enumerators.
2227    if (const EnumType *ET = dyn_cast<EnumType>(T)) {
2228      EnumDecl *Enum = ET->getDecl();
2229      if (!Enum->isDefinition())
2230        return IntRange(C.getIntWidth(QualType(T, 0)), false);
2231
2232      unsigned NumPositive = Enum->getNumPositiveBits();
2233      unsigned NumNegative = Enum->getNumNegativeBits();
2234
2235      return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0);
2236    }
2237
2238    const BuiltinType *BT = cast<BuiltinType>(T);
2239    assert(BT->isInteger());
2240
2241    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2242  }
2243
2244  /// Returns the "target" range of a canonical integral type, i.e.
2245  /// the range of values expressible in the type.
2246  ///
2247  /// This matches forValueOfCanonicalType except that enums have the
2248  /// full range of their type, not the range of their enumerators.
2249  static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) {
2250    assert(T->isCanonicalUnqualified());
2251
2252    if (const VectorType *VT = dyn_cast<VectorType>(T))
2253      T = VT->getElementType().getTypePtr();
2254    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2255      T = CT->getElementType().getTypePtr();
2256    if (const EnumType *ET = dyn_cast<EnumType>(T))
2257      T = ET->getDecl()->getIntegerType().getTypePtr();
2258
2259    const BuiltinType *BT = cast<BuiltinType>(T);
2260    assert(BT->isInteger());
2261
2262    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2263  }
2264
2265  /// Returns the supremum of two ranges: i.e. their conservative merge.
2266  static IntRange join(IntRange L, IntRange R) {
2267    return IntRange(std::max(L.Width, R.Width),
2268                    L.NonNegative && R.NonNegative);
2269  }
2270
2271  /// Returns the infinum of two ranges: i.e. their aggressive merge.
2272  static IntRange meet(IntRange L, IntRange R) {
2273    return IntRange(std::min(L.Width, R.Width),
2274                    L.NonNegative || R.NonNegative);
2275  }
2276};
2277
2278IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) {
2279  if (value.isSigned() && value.isNegative())
2280    return IntRange(value.getMinSignedBits(), false);
2281
2282  if (value.getBitWidth() > MaxWidth)
2283    value = value.trunc(MaxWidth);
2284
2285  // isNonNegative() just checks the sign bit without considering
2286  // signedness.
2287  return IntRange(value.getActiveBits(), true);
2288}
2289
2290IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty,
2291                       unsigned MaxWidth) {
2292  if (result.isInt())
2293    return GetValueRange(C, result.getInt(), MaxWidth);
2294
2295  if (result.isVector()) {
2296    IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth);
2297    for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) {
2298      IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth);
2299      R = IntRange::join(R, El);
2300    }
2301    return R;
2302  }
2303
2304  if (result.isComplexInt()) {
2305    IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth);
2306    IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth);
2307    return IntRange::join(R, I);
2308  }
2309
2310  // This can happen with lossless casts to intptr_t of "based" lvalues.
2311  // Assume it might use arbitrary bits.
2312  // FIXME: The only reason we need to pass the type in here is to get
2313  // the sign right on this one case.  It would be nice if APValue
2314  // preserved this.
2315  assert(result.isLValue());
2316  return IntRange(MaxWidth, Ty->isUnsignedIntegerType());
2317}
2318
2319/// Pseudo-evaluate the given integer expression, estimating the
2320/// range of values it might take.
2321///
2322/// \param MaxWidth - the width to which the value will be truncated
2323IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) {
2324  E = E->IgnoreParens();
2325
2326  // Try a full evaluation first.
2327  Expr::EvalResult result;
2328  if (E->Evaluate(result, C))
2329    return GetValueRange(C, result.Val, E->getType(), MaxWidth);
2330
2331  // I think we only want to look through implicit casts here; if the
2332  // user has an explicit widening cast, we should treat the value as
2333  // being of the new, wider type.
2334  if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) {
2335    if (CE->getCastKind() == CK_NoOp)
2336      return GetExprRange(C, CE->getSubExpr(), MaxWidth);
2337
2338    IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType());
2339
2340    bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast);
2341
2342    // Assume that non-integer casts can span the full range of the type.
2343    if (!isIntegerCast)
2344      return OutputTypeRange;
2345
2346    IntRange SubRange
2347      = GetExprRange(C, CE->getSubExpr(),
2348                     std::min(MaxWidth, OutputTypeRange.Width));
2349
2350    // Bail out if the subexpr's range is as wide as the cast type.
2351    if (SubRange.Width >= OutputTypeRange.Width)
2352      return OutputTypeRange;
2353
2354    // Otherwise, we take the smaller width, and we're non-negative if
2355    // either the output type or the subexpr is.
2356    return IntRange(SubRange.Width,
2357                    SubRange.NonNegative || OutputTypeRange.NonNegative);
2358  }
2359
2360  if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
2361    // If we can fold the condition, just take that operand.
2362    bool CondResult;
2363    if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C))
2364      return GetExprRange(C, CondResult ? CO->getTrueExpr()
2365                                        : CO->getFalseExpr(),
2366                          MaxWidth);
2367
2368    // Otherwise, conservatively merge.
2369    IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth);
2370    IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth);
2371    return IntRange::join(L, R);
2372  }
2373
2374  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
2375    switch (BO->getOpcode()) {
2376
2377    // Boolean-valued operations are single-bit and positive.
2378    case BO_LAnd:
2379    case BO_LOr:
2380    case BO_LT:
2381    case BO_GT:
2382    case BO_LE:
2383    case BO_GE:
2384    case BO_EQ:
2385    case BO_NE:
2386      return IntRange::forBoolType();
2387
2388    // The type of these compound assignments is the type of the LHS,
2389    // so the RHS is not necessarily an integer.
2390    case BO_MulAssign:
2391    case BO_DivAssign:
2392    case BO_RemAssign:
2393    case BO_AddAssign:
2394    case BO_SubAssign:
2395      return IntRange::forValueOfType(C, E->getType());
2396
2397    // Operations with opaque sources are black-listed.
2398    case BO_PtrMemD:
2399    case BO_PtrMemI:
2400      return IntRange::forValueOfType(C, E->getType());
2401
2402    // Bitwise-and uses the *infinum* of the two source ranges.
2403    case BO_And:
2404    case BO_AndAssign:
2405      return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth),
2406                            GetExprRange(C, BO->getRHS(), MaxWidth));
2407
2408    // Left shift gets black-listed based on a judgement call.
2409    case BO_Shl:
2410      // ...except that we want to treat '1 << (blah)' as logically
2411      // positive.  It's an important idiom.
2412      if (IntegerLiteral *I
2413            = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) {
2414        if (I->getValue() == 1) {
2415          IntRange R = IntRange::forValueOfType(C, E->getType());
2416          return IntRange(R.Width, /*NonNegative*/ true);
2417        }
2418      }
2419      // fallthrough
2420
2421    case BO_ShlAssign:
2422      return IntRange::forValueOfType(C, E->getType());
2423
2424    // Right shift by a constant can narrow its left argument.
2425    case BO_Shr:
2426    case BO_ShrAssign: {
2427      IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2428
2429      // If the shift amount is a positive constant, drop the width by
2430      // that much.
2431      llvm::APSInt shift;
2432      if (BO->getRHS()->isIntegerConstantExpr(shift, C) &&
2433          shift.isNonNegative()) {
2434        unsigned zext = shift.getZExtValue();
2435        if (zext >= L.Width)
2436          L.Width = (L.NonNegative ? 0 : 1);
2437        else
2438          L.Width -= zext;
2439      }
2440
2441      return L;
2442    }
2443
2444    // Comma acts as its right operand.
2445    case BO_Comma:
2446      return GetExprRange(C, BO->getRHS(), MaxWidth);
2447
2448    // Black-list pointer subtractions.
2449    case BO_Sub:
2450      if (BO->getLHS()->getType()->isPointerType())
2451        return IntRange::forValueOfType(C, E->getType());
2452      // fallthrough
2453
2454    default:
2455      break;
2456    }
2457
2458    // Treat every other operator as if it were closed on the
2459    // narrowest type that encompasses both operands.
2460    IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2461    IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth);
2462    return IntRange::join(L, R);
2463  }
2464
2465  if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
2466    switch (UO->getOpcode()) {
2467    // Boolean-valued operations are white-listed.
2468    case UO_LNot:
2469      return IntRange::forBoolType();
2470
2471    // Operations with opaque sources are black-listed.
2472    case UO_Deref:
2473    case UO_AddrOf: // should be impossible
2474      return IntRange::forValueOfType(C, E->getType());
2475
2476    default:
2477      return GetExprRange(C, UO->getSubExpr(), MaxWidth);
2478    }
2479  }
2480
2481  if (dyn_cast<OffsetOfExpr>(E)) {
2482    IntRange::forValueOfType(C, E->getType());
2483  }
2484
2485  FieldDecl *BitField = E->getBitField();
2486  if (BitField) {
2487    llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C);
2488    unsigned BitWidth = BitWidthAP.getZExtValue();
2489
2490    return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType());
2491  }
2492
2493  return IntRange::forValueOfType(C, E->getType());
2494}
2495
2496IntRange GetExprRange(ASTContext &C, Expr *E) {
2497  return GetExprRange(C, E, C.getIntWidth(E->getType()));
2498}
2499
2500/// Checks whether the given value, which currently has the given
2501/// source semantics, has the same value when coerced through the
2502/// target semantics.
2503bool IsSameFloatAfterCast(const llvm::APFloat &value,
2504                          const llvm::fltSemantics &Src,
2505                          const llvm::fltSemantics &Tgt) {
2506  llvm::APFloat truncated = value;
2507
2508  bool ignored;
2509  truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored);
2510  truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored);
2511
2512  return truncated.bitwiseIsEqual(value);
2513}
2514
2515/// Checks whether the given value, which currently has the given
2516/// source semantics, has the same value when coerced through the
2517/// target semantics.
2518///
2519/// The value might be a vector of floats (or a complex number).
2520bool IsSameFloatAfterCast(const APValue &value,
2521                          const llvm::fltSemantics &Src,
2522                          const llvm::fltSemantics &Tgt) {
2523  if (value.isFloat())
2524    return IsSameFloatAfterCast(value.getFloat(), Src, Tgt);
2525
2526  if (value.isVector()) {
2527    for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i)
2528      if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt))
2529        return false;
2530    return true;
2531  }
2532
2533  assert(value.isComplexFloat());
2534  return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) &&
2535          IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt));
2536}
2537
2538void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC);
2539
2540static bool IsZero(Sema &S, Expr *E) {
2541  // Suppress cases where we are comparing against an enum constant.
2542  if (const DeclRefExpr *DR =
2543      dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts()))
2544    if (isa<EnumConstantDecl>(DR->getDecl()))
2545      return false;
2546
2547  // Suppress cases where the '0' value is expanded from a macro.
2548  if (E->getLocStart().isMacroID())
2549    return false;
2550
2551  llvm::APSInt Value;
2552  return E->isIntegerConstantExpr(Value, S.Context) && Value == 0;
2553}
2554
2555static bool HasEnumType(Expr *E) {
2556  // Strip off implicit integral promotions.
2557  while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
2558    if (ICE->getCastKind() != CK_IntegralCast &&
2559        ICE->getCastKind() != CK_NoOp)
2560      break;
2561    E = ICE->getSubExpr();
2562  }
2563
2564  return E->getType()->isEnumeralType();
2565}
2566
2567void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) {
2568  BinaryOperatorKind op = E->getOpcode();
2569  if (E->isValueDependent())
2570    return;
2571
2572  if (op == BO_LT && IsZero(S, E->getRHS())) {
2573    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2574      << "< 0" << "false" << HasEnumType(E->getLHS())
2575      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2576  } else if (op == BO_GE && IsZero(S, E->getRHS())) {
2577    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2578      << ">= 0" << "true" << HasEnumType(E->getLHS())
2579      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2580  } else if (op == BO_GT && IsZero(S, E->getLHS())) {
2581    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
2582      << "0 >" << "false" << HasEnumType(E->getRHS())
2583      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2584  } else if (op == BO_LE && IsZero(S, E->getLHS())) {
2585    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
2586      << "0 <=" << "true" << HasEnumType(E->getRHS())
2587      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2588  }
2589}
2590
2591/// Analyze the operands of the given comparison.  Implements the
2592/// fallback case from AnalyzeComparison.
2593void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) {
2594  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
2595  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
2596}
2597
2598/// \brief Implements -Wsign-compare.
2599///
2600/// \param lex the left-hand expression
2601/// \param rex the right-hand expression
2602/// \param OpLoc the location of the joining operator
2603/// \param BinOpc binary opcode or 0
2604void AnalyzeComparison(Sema &S, BinaryOperator *E) {
2605  // The type the comparison is being performed in.
2606  QualType T = E->getLHS()->getType();
2607  assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType())
2608         && "comparison with mismatched types");
2609
2610  // We don't do anything special if this isn't an unsigned integral
2611  // comparison:  we're only interested in integral comparisons, and
2612  // signed comparisons only happen in cases we don't care to warn about.
2613  //
2614  // We also don't care about value-dependent expressions or expressions
2615  // whose result is a constant.
2616  if (!T->hasUnsignedIntegerRepresentation()
2617      || E->isValueDependent() || E->isIntegerConstantExpr(S.Context))
2618    return AnalyzeImpConvsInComparison(S, E);
2619
2620  Expr *lex = E->getLHS()->IgnoreParenImpCasts();
2621  Expr *rex = E->getRHS()->IgnoreParenImpCasts();
2622
2623  // Check to see if one of the (unmodified) operands is of different
2624  // signedness.
2625  Expr *signedOperand, *unsignedOperand;
2626  if (lex->getType()->hasSignedIntegerRepresentation()) {
2627    assert(!rex->getType()->hasSignedIntegerRepresentation() &&
2628           "unsigned comparison between two signed integer expressions?");
2629    signedOperand = lex;
2630    unsignedOperand = rex;
2631  } else if (rex->getType()->hasSignedIntegerRepresentation()) {
2632    signedOperand = rex;
2633    unsignedOperand = lex;
2634  } else {
2635    CheckTrivialUnsignedComparison(S, E);
2636    return AnalyzeImpConvsInComparison(S, E);
2637  }
2638
2639  // Otherwise, calculate the effective range of the signed operand.
2640  IntRange signedRange = GetExprRange(S.Context, signedOperand);
2641
2642  // Go ahead and analyze implicit conversions in the operands.  Note
2643  // that we skip the implicit conversions on both sides.
2644  AnalyzeImplicitConversions(S, lex, E->getOperatorLoc());
2645  AnalyzeImplicitConversions(S, rex, E->getOperatorLoc());
2646
2647  // If the signed range is non-negative, -Wsign-compare won't fire,
2648  // but we should still check for comparisons which are always true
2649  // or false.
2650  if (signedRange.NonNegative)
2651    return CheckTrivialUnsignedComparison(S, E);
2652
2653  // For (in)equality comparisons, if the unsigned operand is a
2654  // constant which cannot collide with a overflowed signed operand,
2655  // then reinterpreting the signed operand as unsigned will not
2656  // change the result of the comparison.
2657  if (E->isEqualityOp()) {
2658    unsigned comparisonWidth = S.Context.getIntWidth(T);
2659    IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand);
2660
2661    // We should never be unable to prove that the unsigned operand is
2662    // non-negative.
2663    assert(unsignedRange.NonNegative && "unsigned range includes negative?");
2664
2665    if (unsignedRange.Width < comparisonWidth)
2666      return;
2667  }
2668
2669  S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison)
2670    << lex->getType() << rex->getType()
2671    << lex->getSourceRange() << rex->getSourceRange();
2672}
2673
2674/// Analyzes an attempt to assign the given value to a bitfield.
2675///
2676/// Returns true if there was something fishy about the attempt.
2677bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init,
2678                               SourceLocation InitLoc) {
2679  assert(Bitfield->isBitField());
2680  if (Bitfield->isInvalidDecl())
2681    return false;
2682
2683  // White-list bool bitfields.
2684  if (Bitfield->getType()->isBooleanType())
2685    return false;
2686
2687  // Ignore value- or type-dependent expressions.
2688  if (Bitfield->getBitWidth()->isValueDependent() ||
2689      Bitfield->getBitWidth()->isTypeDependent() ||
2690      Init->isValueDependent() ||
2691      Init->isTypeDependent())
2692    return false;
2693
2694  Expr *OriginalInit = Init->IgnoreParenImpCasts();
2695
2696  llvm::APSInt Width(32);
2697  Expr::EvalResult InitValue;
2698  if (!Bitfield->getBitWidth()->isIntegerConstantExpr(Width, S.Context) ||
2699      !OriginalInit->Evaluate(InitValue, S.Context) ||
2700      !InitValue.Val.isInt())
2701    return false;
2702
2703  const llvm::APSInt &Value = InitValue.Val.getInt();
2704  unsigned OriginalWidth = Value.getBitWidth();
2705  unsigned FieldWidth = Width.getZExtValue();
2706
2707  if (OriginalWidth <= FieldWidth)
2708    return false;
2709
2710  llvm::APSInt TruncatedValue = Value.trunc(FieldWidth);
2711
2712  // It's fairly common to write values into signed bitfields
2713  // that, if sign-extended, would end up becoming a different
2714  // value.  We don't want to warn about that.
2715  if (Value.isSigned() && Value.isNegative())
2716    TruncatedValue = TruncatedValue.sext(OriginalWidth);
2717  else
2718    TruncatedValue = TruncatedValue.zext(OriginalWidth);
2719
2720  if (Value == TruncatedValue)
2721    return false;
2722
2723  std::string PrettyValue = Value.toString(10);
2724  std::string PrettyTrunc = TruncatedValue.toString(10);
2725
2726  S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant)
2727    << PrettyValue << PrettyTrunc << OriginalInit->getType()
2728    << Init->getSourceRange();
2729
2730  return true;
2731}
2732
2733/// Analyze the given simple or compound assignment for warning-worthy
2734/// operations.
2735void AnalyzeAssignment(Sema &S, BinaryOperator *E) {
2736  // Just recurse on the LHS.
2737  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
2738
2739  // We want to recurse on the RHS as normal unless we're assigning to
2740  // a bitfield.
2741  if (FieldDecl *Bitfield = E->getLHS()->getBitField()) {
2742    if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(),
2743                                  E->getOperatorLoc())) {
2744      // Recurse, ignoring any implicit conversions on the RHS.
2745      return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(),
2746                                        E->getOperatorLoc());
2747    }
2748  }
2749
2750  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
2751}
2752
2753/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
2754void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T,
2755                     SourceLocation CContext, unsigned diag) {
2756  S.Diag(E->getExprLoc(), diag)
2757    << SourceType << T << E->getSourceRange() << SourceRange(CContext);
2758}
2759
2760/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
2761void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext,
2762                     unsigned diag) {
2763  DiagnoseImpCast(S, E, E->getType(), T, CContext, diag);
2764}
2765
2766/// Diagnose an implicit cast from a literal expression. Also attemps to supply
2767/// fixit hints when the cast wouldn't lose information to simply write the
2768/// expression with the expected type.
2769void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T,
2770                                    SourceLocation CContext) {
2771  // Emit the primary warning first, then try to emit a fixit hint note if
2772  // reasonable.
2773  S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer)
2774    << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext);
2775
2776  const llvm::APFloat &Value = FL->getValue();
2777
2778  // Don't attempt to fix PPC double double literals.
2779  if (&Value.getSemantics() == &llvm::APFloat::PPCDoubleDouble)
2780    return;
2781
2782  // Try to convert this exactly to an 64-bit integer. FIXME: It would be
2783  // nice to support arbitrarily large integers here.
2784  bool isExact = false;
2785  uint64_t IntegerPart;
2786  if (Value.convertToInteger(&IntegerPart, 64, /*isSigned=*/true,
2787                             llvm::APFloat::rmTowardZero, &isExact)
2788      != llvm::APFloat::opOK || !isExact)
2789    return;
2790
2791  llvm::APInt IntegerValue(64, IntegerPart, /*isSigned=*/true);
2792
2793  std::string LiteralValue = IntegerValue.toString(10, /*isSigned=*/true);
2794  S.Diag(FL->getExprLoc(), diag::note_fix_integral_float_as_integer)
2795    << FixItHint::CreateReplacement(FL->getSourceRange(), LiteralValue);
2796}
2797
2798std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) {
2799  if (!Range.Width) return "0";
2800
2801  llvm::APSInt ValueInRange = Value;
2802  ValueInRange.setIsSigned(!Range.NonNegative);
2803  ValueInRange = ValueInRange.trunc(Range.Width);
2804  return ValueInRange.toString(10);
2805}
2806
2807static bool isFromSystemMacro(Sema &S, SourceLocation loc) {
2808  SourceManager &smgr = S.Context.getSourceManager();
2809  return loc.isMacroID() && smgr.isInSystemHeader(smgr.getSpellingLoc(loc));
2810}
2811
2812void CheckImplicitConversion(Sema &S, Expr *E, QualType T,
2813                             SourceLocation CC, bool *ICContext = 0) {
2814  if (E->isTypeDependent() || E->isValueDependent()) return;
2815
2816  const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr();
2817  const Type *Target = S.Context.getCanonicalType(T).getTypePtr();
2818  if (Source == Target) return;
2819  if (Target->isDependentType()) return;
2820
2821  // If the conversion context location is invalid don't complain.
2822  // We also don't want to emit a warning if the issue occurs from the
2823  // instantiation of a system macro.  The problem is that 'getSpellingLoc()'
2824  // is slow, so we delay this check as long as possible.  Once we detect
2825  // we are in that scenario, we just return.
2826  if (CC.isInvalid())
2827    return;
2828
2829  // Never diagnose implicit casts to bool.
2830  if (Target->isSpecificBuiltinType(BuiltinType::Bool))
2831    return;
2832
2833  // Strip vector types.
2834  if (isa<VectorType>(Source)) {
2835    if (!isa<VectorType>(Target)) {
2836      if (isFromSystemMacro(S, CC))
2837        return;
2838      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar);
2839    }
2840
2841    Source = cast<VectorType>(Source)->getElementType().getTypePtr();
2842    Target = cast<VectorType>(Target)->getElementType().getTypePtr();
2843  }
2844
2845  // Strip complex types.
2846  if (isa<ComplexType>(Source)) {
2847    if (!isa<ComplexType>(Target)) {
2848      if (isFromSystemMacro(S, CC))
2849        return;
2850
2851      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar);
2852    }
2853
2854    Source = cast<ComplexType>(Source)->getElementType().getTypePtr();
2855    Target = cast<ComplexType>(Target)->getElementType().getTypePtr();
2856  }
2857
2858  const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source);
2859  const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target);
2860
2861  // If the source is floating point...
2862  if (SourceBT && SourceBT->isFloatingPoint()) {
2863    // ...and the target is floating point...
2864    if (TargetBT && TargetBT->isFloatingPoint()) {
2865      // ...then warn if we're dropping FP rank.
2866
2867      // Builtin FP kinds are ordered by increasing FP rank.
2868      if (SourceBT->getKind() > TargetBT->getKind()) {
2869        // Don't warn about float constants that are precisely
2870        // representable in the target type.
2871        Expr::EvalResult result;
2872        if (E->Evaluate(result, S.Context)) {
2873          // Value might be a float, a float vector, or a float complex.
2874          if (IsSameFloatAfterCast(result.Val,
2875                   S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)),
2876                   S.Context.getFloatTypeSemantics(QualType(SourceBT, 0))))
2877            return;
2878        }
2879
2880        if (isFromSystemMacro(S, CC))
2881          return;
2882
2883        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision);
2884      }
2885      return;
2886    }
2887
2888    // If the target is integral, always warn.
2889    if ((TargetBT && TargetBT->isInteger())) {
2890      if (isFromSystemMacro(S, CC))
2891        return;
2892
2893      Expr *InnerE = E->IgnoreParenImpCasts();
2894      if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) {
2895        DiagnoseFloatingLiteralImpCast(S, FL, T, CC);
2896      } else {
2897        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer);
2898      }
2899    }
2900
2901    return;
2902  }
2903
2904  if (!Source->isIntegerType() || !Target->isIntegerType())
2905    return;
2906
2907  IntRange SourceRange = GetExprRange(S.Context, E);
2908  IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target);
2909
2910  if (SourceRange.Width > TargetRange.Width) {
2911    // If the source is a constant, use a default-on diagnostic.
2912    // TODO: this should happen for bitfield stores, too.
2913    llvm::APSInt Value(32);
2914    if (E->isIntegerConstantExpr(Value, S.Context)) {
2915      if (isFromSystemMacro(S, CC))
2916        return;
2917
2918      std::string PrettySourceValue = Value.toString(10);
2919      std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange);
2920
2921      S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant)
2922        << PrettySourceValue << PrettyTargetValue
2923        << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC);
2924      return;
2925    }
2926
2927    // People want to build with -Wshorten-64-to-32 and not -Wconversion
2928    // and by god we'll let them.
2929
2930    if (isFromSystemMacro(S, CC))
2931      return;
2932
2933    if (SourceRange.Width == 64 && TargetRange.Width == 32)
2934      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32);
2935    return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision);
2936  }
2937
2938  if ((TargetRange.NonNegative && !SourceRange.NonNegative) ||
2939      (!TargetRange.NonNegative && SourceRange.NonNegative &&
2940       SourceRange.Width == TargetRange.Width)) {
2941
2942    if (isFromSystemMacro(S, CC))
2943      return;
2944
2945    unsigned DiagID = diag::warn_impcast_integer_sign;
2946
2947    // Traditionally, gcc has warned about this under -Wsign-compare.
2948    // We also want to warn about it in -Wconversion.
2949    // So if -Wconversion is off, use a completely identical diagnostic
2950    // in the sign-compare group.
2951    // The conditional-checking code will
2952    if (ICContext) {
2953      DiagID = diag::warn_impcast_integer_sign_conditional;
2954      *ICContext = true;
2955    }
2956
2957    return DiagnoseImpCast(S, E, T, CC, DiagID);
2958  }
2959
2960  // Diagnose conversions between different enumeration types.
2961  // In C, we pretend that the type of an EnumConstantDecl is its enumeration
2962  // type, to give us better diagnostics.
2963  QualType SourceType = E->getType();
2964  if (!S.getLangOptions().CPlusPlus) {
2965    if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
2966      if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) {
2967        EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext());
2968        SourceType = S.Context.getTypeDeclType(Enum);
2969        Source = S.Context.getCanonicalType(SourceType).getTypePtr();
2970      }
2971  }
2972
2973  if (const EnumType *SourceEnum = Source->getAs<EnumType>())
2974    if (const EnumType *TargetEnum = Target->getAs<EnumType>())
2975      if ((SourceEnum->getDecl()->getIdentifier() ||
2976           SourceEnum->getDecl()->getTypedefForAnonDecl()) &&
2977          (TargetEnum->getDecl()->getIdentifier() ||
2978           TargetEnum->getDecl()->getTypedefForAnonDecl()) &&
2979          SourceEnum != TargetEnum) {
2980        if (isFromSystemMacro(S, CC))
2981          return;
2982
2983        return DiagnoseImpCast(S, E, SourceType, T, CC,
2984                               diag::warn_impcast_different_enum_types);
2985      }
2986
2987  return;
2988}
2989
2990void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T);
2991
2992void CheckConditionalOperand(Sema &S, Expr *E, QualType T,
2993                             SourceLocation CC, bool &ICContext) {
2994  E = E->IgnoreParenImpCasts();
2995
2996  if (isa<ConditionalOperator>(E))
2997    return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T);
2998
2999  AnalyzeImplicitConversions(S, E, CC);
3000  if (E->getType() != T)
3001    return CheckImplicitConversion(S, E, T, CC, &ICContext);
3002  return;
3003}
3004
3005void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) {
3006  SourceLocation CC = E->getQuestionLoc();
3007
3008  AnalyzeImplicitConversions(S, E->getCond(), CC);
3009
3010  bool Suspicious = false;
3011  CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious);
3012  CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious);
3013
3014  // If -Wconversion would have warned about either of the candidates
3015  // for a signedness conversion to the context type...
3016  if (!Suspicious) return;
3017
3018  // ...but it's currently ignored...
3019  if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional,
3020                                 CC))
3021    return;
3022
3023  // ...and -Wsign-compare isn't...
3024  if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional, CC))
3025    return;
3026
3027  // ...then check whether it would have warned about either of the
3028  // candidates for a signedness conversion to the condition type.
3029  if (E->getType() != T) {
3030    Suspicious = false;
3031    CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(),
3032                            E->getType(), CC, &Suspicious);
3033    if (!Suspicious)
3034      CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(),
3035                              E->getType(), CC, &Suspicious);
3036    if (!Suspicious)
3037      return;
3038  }
3039
3040  // If so, emit a diagnostic under -Wsign-compare.
3041  Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts();
3042  Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts();
3043  S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional)
3044    << lex->getType() << rex->getType()
3045    << lex->getSourceRange() << rex->getSourceRange();
3046}
3047
3048/// AnalyzeImplicitConversions - Find and report any interesting
3049/// implicit conversions in the given expression.  There are a couple
3050/// of competing diagnostics here, -Wconversion and -Wsign-compare.
3051void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) {
3052  QualType T = OrigE->getType();
3053  Expr *E = OrigE->IgnoreParenImpCasts();
3054
3055  // For conditional operators, we analyze the arguments as if they
3056  // were being fed directly into the output.
3057  if (isa<ConditionalOperator>(E)) {
3058    ConditionalOperator *CO = cast<ConditionalOperator>(E);
3059    CheckConditionalOperator(S, CO, T);
3060    return;
3061  }
3062
3063  // Go ahead and check any implicit conversions we might have skipped.
3064  // The non-canonical typecheck is just an optimization;
3065  // CheckImplicitConversion will filter out dead implicit conversions.
3066  if (E->getType() != T)
3067    CheckImplicitConversion(S, E, T, CC);
3068
3069  // Now continue drilling into this expression.
3070
3071  // Skip past explicit casts.
3072  if (isa<ExplicitCastExpr>(E)) {
3073    E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts();
3074    return AnalyzeImplicitConversions(S, E, CC);
3075  }
3076
3077  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
3078    // Do a somewhat different check with comparison operators.
3079    if (BO->isComparisonOp())
3080      return AnalyzeComparison(S, BO);
3081
3082    // And with assignments and compound assignments.
3083    if (BO->isAssignmentOp())
3084      return AnalyzeAssignment(S, BO);
3085  }
3086
3087  // These break the otherwise-useful invariant below.  Fortunately,
3088  // we don't really need to recurse into them, because any internal
3089  // expressions should have been analyzed already when they were
3090  // built into statements.
3091  if (isa<StmtExpr>(E)) return;
3092
3093  // Don't descend into unevaluated contexts.
3094  if (isa<UnaryExprOrTypeTraitExpr>(E)) return;
3095
3096  // Now just recurse over the expression's children.
3097  CC = E->getExprLoc();
3098  for (Stmt::child_range I = E->children(); I; ++I)
3099    AnalyzeImplicitConversions(S, cast<Expr>(*I), CC);
3100}
3101
3102} // end anonymous namespace
3103
3104/// Diagnoses "dangerous" implicit conversions within the given
3105/// expression (which is a full expression).  Implements -Wconversion
3106/// and -Wsign-compare.
3107///
3108/// \param CC the "context" location of the implicit conversion, i.e.
3109///   the most location of the syntactic entity requiring the implicit
3110///   conversion
3111void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) {
3112  // Don't diagnose in unevaluated contexts.
3113  if (ExprEvalContexts.back().Context == Sema::Unevaluated)
3114    return;
3115
3116  // Don't diagnose for value- or type-dependent expressions.
3117  if (E->isTypeDependent() || E->isValueDependent())
3118    return;
3119
3120  // This is not the right CC for (e.g.) a variable initialization.
3121  AnalyzeImplicitConversions(*this, E, CC);
3122}
3123
3124void Sema::CheckBitFieldInitialization(SourceLocation InitLoc,
3125                                       FieldDecl *BitField,
3126                                       Expr *Init) {
3127  (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc);
3128}
3129
3130/// CheckParmsForFunctionDef - Check that the parameters of the given
3131/// function are appropriate for the definition of a function. This
3132/// takes care of any checks that cannot be performed on the
3133/// declaration itself, e.g., that the types of each of the function
3134/// parameters are complete.
3135bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd,
3136                                    bool CheckParameterNames) {
3137  bool HasInvalidParm = false;
3138  for (; P != PEnd; ++P) {
3139    ParmVarDecl *Param = *P;
3140
3141    // C99 6.7.5.3p4: the parameters in a parameter type list in a
3142    // function declarator that is part of a function definition of
3143    // that function shall not have incomplete type.
3144    //
3145    // This is also C++ [dcl.fct]p6.
3146    if (!Param->isInvalidDecl() &&
3147        RequireCompleteType(Param->getLocation(), Param->getType(),
3148                               diag::err_typecheck_decl_incomplete_type)) {
3149      Param->setInvalidDecl();
3150      HasInvalidParm = true;
3151    }
3152
3153    // C99 6.9.1p5: If the declarator includes a parameter type list, the
3154    // declaration of each parameter shall include an identifier.
3155    if (CheckParameterNames &&
3156        Param->getIdentifier() == 0 &&
3157        !Param->isImplicit() &&
3158        !getLangOptions().CPlusPlus)
3159      Diag(Param->getLocation(), diag::err_parameter_name_omitted);
3160
3161    // C99 6.7.5.3p12:
3162    //   If the function declarator is not part of a definition of that
3163    //   function, parameters may have incomplete type and may use the [*]
3164    //   notation in their sequences of declarator specifiers to specify
3165    //   variable length array types.
3166    QualType PType = Param->getOriginalType();
3167    if (const ArrayType *AT = Context.getAsArrayType(PType)) {
3168      if (AT->getSizeModifier() == ArrayType::Star) {
3169        // FIXME: This diagnosic should point the the '[*]' if source-location
3170        // information is added for it.
3171        Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
3172      }
3173    }
3174  }
3175
3176  return HasInvalidParm;
3177}
3178
3179/// CheckCastAlign - Implements -Wcast-align, which warns when a
3180/// pointer cast increases the alignment requirements.
3181void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) {
3182  // This is actually a lot of work to potentially be doing on every
3183  // cast; don't do it if we're ignoring -Wcast_align (as is the default).
3184  if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align,
3185                                          TRange.getBegin())
3186        == Diagnostic::Ignored)
3187    return;
3188
3189  // Ignore dependent types.
3190  if (T->isDependentType() || Op->getType()->isDependentType())
3191    return;
3192
3193  // Require that the destination be a pointer type.
3194  const PointerType *DestPtr = T->getAs<PointerType>();
3195  if (!DestPtr) return;
3196
3197  // If the destination has alignment 1, we're done.
3198  QualType DestPointee = DestPtr->getPointeeType();
3199  if (DestPointee->isIncompleteType()) return;
3200  CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee);
3201  if (DestAlign.isOne()) return;
3202
3203  // Require that the source be a pointer type.
3204  const PointerType *SrcPtr = Op->getType()->getAs<PointerType>();
3205  if (!SrcPtr) return;
3206  QualType SrcPointee = SrcPtr->getPointeeType();
3207
3208  // Whitelist casts from cv void*.  We already implicitly
3209  // whitelisted casts to cv void*, since they have alignment 1.
3210  // Also whitelist casts involving incomplete types, which implicitly
3211  // includes 'void'.
3212  if (SrcPointee->isIncompleteType()) return;
3213
3214  CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee);
3215  if (SrcAlign >= DestAlign) return;
3216
3217  Diag(TRange.getBegin(), diag::warn_cast_align)
3218    << Op->getType() << T
3219    << static_cast<unsigned>(SrcAlign.getQuantity())
3220    << static_cast<unsigned>(DestAlign.getQuantity())
3221    << TRange << Op->getSourceRange();
3222}
3223
3224static void CheckArrayAccess_Check(Sema &S,
3225                                   const clang::ArraySubscriptExpr *E) {
3226  const Expr *BaseExpr = E->getBase()->IgnoreParenImpCasts();
3227  const ConstantArrayType *ArrayTy =
3228    S.Context.getAsConstantArrayType(BaseExpr->getType());
3229  if (!ArrayTy)
3230    return;
3231
3232  const Expr *IndexExpr = E->getIdx();
3233  if (IndexExpr->isValueDependent())
3234    return;
3235  llvm::APSInt index;
3236  if (!IndexExpr->isIntegerConstantExpr(index, S.Context))
3237    return;
3238
3239  if (index.isUnsigned() || !index.isNegative()) {
3240    llvm::APInt size = ArrayTy->getSize();
3241    if (!size.isStrictlyPositive())
3242      return;
3243    if (size.getBitWidth() > index.getBitWidth())
3244      index = index.sext(size.getBitWidth());
3245    else if (size.getBitWidth() < index.getBitWidth())
3246      size = size.sext(index.getBitWidth());
3247
3248    if (index.slt(size))
3249      return;
3250
3251    S.DiagRuntimeBehavior(E->getBase()->getLocStart(), BaseExpr,
3252                          S.PDiag(diag::warn_array_index_exceeds_bounds)
3253                            << index.toString(10, true)
3254                            << size.toString(10, true)
3255                            << IndexExpr->getSourceRange());
3256  } else {
3257    S.DiagRuntimeBehavior(E->getBase()->getLocStart(), BaseExpr,
3258                          S.PDiag(diag::warn_array_index_precedes_bounds)
3259                            << index.toString(10, true)
3260                            << IndexExpr->getSourceRange());
3261  }
3262
3263  const NamedDecl *ND = NULL;
3264  if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
3265    ND = dyn_cast<NamedDecl>(DRE->getDecl());
3266  if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
3267    ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
3268  if (ND)
3269    S.DiagRuntimeBehavior(ND->getLocStart(), BaseExpr,
3270                          S.PDiag(diag::note_array_index_out_of_bounds)
3271                            << ND->getDeclName());
3272}
3273
3274void Sema::CheckArrayAccess(const Expr *expr) {
3275  while (true)
3276    switch (expr->getStmtClass()) {
3277      case Stmt::ParenExprClass:
3278        expr = cast<ParenExpr>(expr)->getSubExpr();
3279        continue;
3280      case Stmt::ArraySubscriptExprClass:
3281        CheckArrayAccess_Check(*this, cast<ArraySubscriptExpr>(expr));
3282        return;
3283      case Stmt::ConditionalOperatorClass: {
3284        const ConditionalOperator *cond = cast<ConditionalOperator>(expr);
3285        if (const Expr *lhs = cond->getLHS())
3286          CheckArrayAccess(lhs);
3287        if (const Expr *rhs = cond->getRHS())
3288          CheckArrayAccess(rhs);
3289        return;
3290      }
3291      default:
3292        return;
3293    }
3294}
3295