SemaChecking.cpp revision f762905bdefad77f91c7c6782a9c17e6b274d393
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 "Sema.h"
16#include "clang/Analysis/Analyses/FormatString.h"
17#include "clang/AST/ASTContext.h"
18#include "clang/AST/CharUnits.h"
19#include "clang/AST/DeclObjC.h"
20#include "clang/AST/ExprCXX.h"
21#include "clang/AST/ExprObjC.h"
22#include "clang/AST/DeclObjC.h"
23#include "clang/AST/StmtCXX.h"
24#include "clang/AST/StmtObjC.h"
25#include "clang/Lex/LiteralSupport.h"
26#include "clang/Lex/Preprocessor.h"
27#include "llvm/ADT/BitVector.h"
28#include "llvm/ADT/STLExtras.h"
29#include "llvm/ADT/StringExtras.h"
30#include "llvm/Support/raw_ostream.h"
31#include "clang/Basic/TargetBuiltins.h"
32#include "clang/Basic/TargetInfo.h"
33#include <limits>
34using namespace clang;
35
36/// getLocationOfStringLiteralByte - Return a source location that points to the
37/// specified byte of the specified string literal.
38///
39/// Strings are amazingly complex.  They can be formed from multiple tokens and
40/// can have escape sequences in them in addition to the usual trigraph and
41/// escaped newline business.  This routine handles this complexity.
42///
43SourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL,
44                                                    unsigned ByteNo) const {
45  assert(!SL->isWide() && "This doesn't work for wide strings yet");
46
47  // Loop over all of the tokens in this string until we find the one that
48  // contains the byte we're looking for.
49  unsigned TokNo = 0;
50  while (1) {
51    assert(TokNo < SL->getNumConcatenated() && "Invalid byte number!");
52    SourceLocation StrTokLoc = SL->getStrTokenLoc(TokNo);
53
54    // Get the spelling of the string so that we can get the data that makes up
55    // the string literal, not the identifier for the macro it is potentially
56    // expanded through.
57    SourceLocation StrTokSpellingLoc = SourceMgr.getSpellingLoc(StrTokLoc);
58
59    // Re-lex the token to get its length and original spelling.
60    std::pair<FileID, unsigned> LocInfo =
61      SourceMgr.getDecomposedLoc(StrTokSpellingLoc);
62    bool Invalid = false;
63    llvm::StringRef Buffer = SourceMgr.getBufferData(LocInfo.first, &Invalid);
64    if (Invalid)
65      return StrTokSpellingLoc;
66
67    const char *StrData = Buffer.data()+LocInfo.second;
68
69    // Create a langops struct and enable trigraphs.  This is sufficient for
70    // relexing tokens.
71    LangOptions LangOpts;
72    LangOpts.Trigraphs = true;
73
74    // Create a lexer starting at the beginning of this token.
75    Lexer TheLexer(StrTokSpellingLoc, LangOpts, Buffer.begin(), StrData,
76                   Buffer.end());
77    Token TheTok;
78    TheLexer.LexFromRawLexer(TheTok);
79
80    // Use the StringLiteralParser to compute the length of the string in bytes.
81    StringLiteralParser SLP(&TheTok, 1, PP, /*Complain=*/false);
82    unsigned TokNumBytes = SLP.GetStringLength();
83
84    // If the byte is in this token, return the location of the byte.
85    if (ByteNo < TokNumBytes ||
86        (ByteNo == TokNumBytes && TokNo == SL->getNumConcatenated())) {
87      unsigned Offset =
88        StringLiteralParser::getOffsetOfStringByte(TheTok, ByteNo, PP,
89                                                   /*Complain=*/false);
90
91      // Now that we know the offset of the token in the spelling, use the
92      // preprocessor to get the offset in the original source.
93      return PP.AdvanceToTokenCharacter(StrTokLoc, Offset);
94    }
95
96    // Move to the next string token.
97    ++TokNo;
98    ByteNo -= TokNumBytes;
99  }
100}
101
102/// CheckablePrintfAttr - does a function call have a "printf" attribute
103/// and arguments that merit checking?
104bool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) {
105  if (Format->getType() == "printf") return true;
106  if (Format->getType() == "printf0") {
107    // printf0 allows null "format" string; if so don't check format/args
108    unsigned format_idx = Format->getFormatIdx() - 1;
109    // Does the index refer to the implicit object argument?
110    if (isa<CXXMemberCallExpr>(TheCall)) {
111      if (format_idx == 0)
112        return false;
113      --format_idx;
114    }
115    if (format_idx < TheCall->getNumArgs()) {
116      Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts();
117      if (!Format->isNullPointerConstant(Context,
118                                         Expr::NPC_ValueDependentIsNull))
119        return true;
120    }
121  }
122  return false;
123}
124
125Action::OwningExprResult
126Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
127  OwningExprResult TheCallResult(Owned(TheCall));
128
129  switch (BuiltinID) {
130  case Builtin::BI__builtin___CFStringMakeConstantString:
131    assert(TheCall->getNumArgs() == 1 &&
132           "Wrong # arguments to builtin CFStringMakeConstantString");
133    if (CheckObjCString(TheCall->getArg(0)))
134      return ExprError();
135    break;
136  case Builtin::BI__builtin_stdarg_start:
137  case Builtin::BI__builtin_va_start:
138    if (SemaBuiltinVAStart(TheCall))
139      return ExprError();
140    break;
141  case Builtin::BI__builtin_isgreater:
142  case Builtin::BI__builtin_isgreaterequal:
143  case Builtin::BI__builtin_isless:
144  case Builtin::BI__builtin_islessequal:
145  case Builtin::BI__builtin_islessgreater:
146  case Builtin::BI__builtin_isunordered:
147    if (SemaBuiltinUnorderedCompare(TheCall))
148      return ExprError();
149    break;
150  case Builtin::BI__builtin_fpclassify:
151    if (SemaBuiltinFPClassification(TheCall, 6))
152      return ExprError();
153    break;
154  case Builtin::BI__builtin_isfinite:
155  case Builtin::BI__builtin_isinf:
156  case Builtin::BI__builtin_isinf_sign:
157  case Builtin::BI__builtin_isnan:
158  case Builtin::BI__builtin_isnormal:
159    if (SemaBuiltinFPClassification(TheCall, 1))
160      return ExprError();
161    break;
162  case Builtin::BI__builtin_return_address:
163  case Builtin::BI__builtin_frame_address: {
164    llvm::APSInt Result;
165    if (SemaBuiltinConstantArg(TheCall, 0, Result))
166      return ExprError();
167    break;
168  }
169  case Builtin::BI__builtin_eh_return_data_regno: {
170    llvm::APSInt Result;
171    if (SemaBuiltinConstantArg(TheCall, 0, Result))
172      return ExprError();
173    break;
174  }
175  case Builtin::BI__builtin_shufflevector:
176    return SemaBuiltinShuffleVector(TheCall);
177    // TheCall will be freed by the smart pointer here, but that's fine, since
178    // SemaBuiltinShuffleVector guts it, but then doesn't release it.
179  case Builtin::BI__builtin_prefetch:
180    if (SemaBuiltinPrefetch(TheCall))
181      return ExprError();
182    break;
183  case Builtin::BI__builtin_object_size:
184    if (SemaBuiltinObjectSize(TheCall))
185      return ExprError();
186    break;
187  case Builtin::BI__builtin_longjmp:
188    if (SemaBuiltinLongjmp(TheCall))
189      return ExprError();
190    break;
191  case Builtin::BI__sync_fetch_and_add:
192  case Builtin::BI__sync_fetch_and_sub:
193  case Builtin::BI__sync_fetch_and_or:
194  case Builtin::BI__sync_fetch_and_and:
195  case Builtin::BI__sync_fetch_and_xor:
196  case Builtin::BI__sync_add_and_fetch:
197  case Builtin::BI__sync_sub_and_fetch:
198  case Builtin::BI__sync_and_and_fetch:
199  case Builtin::BI__sync_or_and_fetch:
200  case Builtin::BI__sync_xor_and_fetch:
201  case Builtin::BI__sync_val_compare_and_swap:
202  case Builtin::BI__sync_bool_compare_and_swap:
203  case Builtin::BI__sync_lock_test_and_set:
204  case Builtin::BI__sync_lock_release:
205    return SemaBuiltinAtomicOverloaded(move(TheCallResult));
206  }
207
208  // Since the target specific builtins for each arch overlap, only check those
209  // of the arch we are compiling for.
210  if (BuiltinID >= Builtin::FirstTSBuiltin) {
211    switch (Context.Target.getTriple().getArch()) {
212      case llvm::Triple::arm:
213      case llvm::Triple::thumb:
214        if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall))
215          return ExprError();
216        break;
217      case llvm::Triple::x86:
218      case llvm::Triple::x86_64:
219        if (CheckX86BuiltinFunctionCall(BuiltinID, TheCall))
220          return ExprError();
221        break;
222      default:
223        break;
224    }
225  }
226
227  return move(TheCallResult);
228}
229
230bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
231  switch (BuiltinID) {
232  case X86::BI__builtin_ia32_palignr128:
233  case X86::BI__builtin_ia32_palignr: {
234    llvm::APSInt Result;
235    if (SemaBuiltinConstantArg(TheCall, 2, Result))
236      return true;
237    break;
238  }
239  }
240  return false;
241}
242
243// Get the valid immediate range for the specified NEON type code.
244static unsigned RFT(unsigned t, bool shift = false) {
245  bool quad = t & 0x10;
246
247  switch (t & 0x7) {
248    case 0: // i8
249      return shift ? 7 : (8 << (int)quad) - 1;
250    case 1: // i16
251      return shift ? 15 : (4 << (int)quad) - 1;
252    case 2: // i32
253      return shift ? 31 : (2 << (int)quad) - 1;
254    case 3: // i64
255      return shift ? 63 : (1 << (int)quad) - 1;
256    case 4: // f32
257      assert(!shift && "cannot shift float types!");
258      return (2 << (int)quad) - 1;
259    case 5: // poly8
260      assert(!shift && "cannot shift polynomial types!");
261      return (8 << (int)quad) - 1;
262    case 6: // poly16
263      assert(!shift && "cannot shift polynomial types!");
264      return (4 << (int)quad) - 1;
265    case 7: // float16
266      assert(!shift && "cannot shift float types!");
267      return (4 << (int)quad) - 1;
268  }
269  return 0;
270}
271
272bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
273  llvm::APSInt Result;
274
275  unsigned mask = 0;
276  unsigned TV = 0;
277  switch (BuiltinID) {
278#define GET_NEON_OVERLOAD_CHECK
279#include "clang/Basic/arm_neon.inc"
280#undef GET_NEON_OVERLOAD_CHECK
281  }
282
283  // For NEON intrinsics which are overloaded on vector element type, validate
284  // the immediate which specifies which variant to emit.
285  if (mask) {
286    unsigned ArgNo = TheCall->getNumArgs()-1;
287    if (SemaBuiltinConstantArg(TheCall, ArgNo, Result))
288      return true;
289
290    TV = Result.getLimitedValue(32);
291    if ((TV > 31) || (mask & (1 << TV)) == 0)
292      return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code)
293        << TheCall->getArg(ArgNo)->getSourceRange();
294  }
295
296  // For NEON intrinsics which take an immediate value as part of the
297  // instruction, range check them here.
298  unsigned i = 0, l = 0, u = 0;
299  switch (BuiltinID) {
300  default: return false;
301#define GET_NEON_IMMEDIATE_CHECK
302#include "clang/Basic/arm_neon.inc"
303#undef GET_NEON_IMMEDIATE_CHECK
304  };
305
306  // Check that the immediate argument is actually a constant.
307  if (SemaBuiltinConstantArg(TheCall, i, Result))
308    return true;
309
310  // Range check against the upper/lower values for this isntruction.
311  unsigned Val = Result.getZExtValue();
312  if (Val < l || Val > (u + l))
313    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
314      << llvm::utostr(l) << llvm::utostr(u+l)
315      << TheCall->getArg(i)->getSourceRange();
316
317  return false;
318}
319
320/// CheckFunctionCall - Check a direct function call for various correctness
321/// and safety properties not strictly enforced by the C type system.
322bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) {
323  // Get the IdentifierInfo* for the called function.
324  IdentifierInfo *FnInfo = FDecl->getIdentifier();
325
326  // None of the checks below are needed for functions that don't have
327  // simple names (e.g., C++ conversion functions).
328  if (!FnInfo)
329    return false;
330
331  // FIXME: This mechanism should be abstracted to be less fragile and
332  // more efficient. For example, just map function ids to custom
333  // handlers.
334
335  // Printf checking.
336  if (const FormatAttr *Format = FDecl->getAttr<FormatAttr>()) {
337    const bool b = Format->getType() == "scanf";
338    if (b || CheckablePrintfAttr(Format, TheCall)) {
339      bool HasVAListArg = Format->getFirstArg() == 0;
340      CheckPrintfScanfArguments(TheCall, HasVAListArg,
341                                Format->getFormatIdx() - 1,
342                                HasVAListArg ? 0 : Format->getFirstArg() - 1,
343                                !b);
344    }
345  }
346
347  for (const NonNullAttr *NonNull = FDecl->getAttr<NonNullAttr>(); NonNull;
348       NonNull = NonNull->getNext<NonNullAttr>())
349    CheckNonNullArguments(NonNull, TheCall);
350
351  return false;
352}
353
354bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) {
355  // Printf checking.
356  const FormatAttr *Format = NDecl->getAttr<FormatAttr>();
357  if (!Format)
358    return false;
359
360  const VarDecl *V = dyn_cast<VarDecl>(NDecl);
361  if (!V)
362    return false;
363
364  QualType Ty = V->getType();
365  if (!Ty->isBlockPointerType())
366    return false;
367
368  const bool b = Format->getType() == "scanf";
369  if (!b && !CheckablePrintfAttr(Format, TheCall))
370    return false;
371
372  bool HasVAListArg = Format->getFirstArg() == 0;
373  CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1,
374                            HasVAListArg ? 0 : Format->getFirstArg() - 1, !b);
375
376  return false;
377}
378
379/// SemaBuiltinAtomicOverloaded - We have a call to a function like
380/// __sync_fetch_and_add, which is an overloaded function based on the pointer
381/// type of its first argument.  The main ActOnCallExpr routines have already
382/// promoted the types of arguments because all of these calls are prototyped as
383/// void(...).
384///
385/// This function goes through and does final semantic checking for these
386/// builtins,
387Sema::OwningExprResult
388Sema::SemaBuiltinAtomicOverloaded(OwningExprResult TheCallResult) {
389  CallExpr *TheCall = (CallExpr *)TheCallResult.get();
390  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
391  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
392
393  // Ensure that we have at least one argument to do type inference from.
394  if (TheCall->getNumArgs() < 1) {
395    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
396      << 0 << 1 << TheCall->getNumArgs()
397      << TheCall->getCallee()->getSourceRange();
398    return ExprError();
399  }
400
401  // Inspect the first argument of the atomic builtin.  This should always be
402  // a pointer type, whose element is an integral scalar or pointer type.
403  // Because it is a pointer type, we don't have to worry about any implicit
404  // casts here.
405  // FIXME: We don't allow floating point scalars as input.
406  Expr *FirstArg = TheCall->getArg(0);
407  if (!FirstArg->getType()->isPointerType()) {
408    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
409      << FirstArg->getType() << FirstArg->getSourceRange();
410    return ExprError();
411  }
412
413  QualType ValType =
414    FirstArg->getType()->getAs<PointerType>()->getPointeeType();
415  if (!ValType->isIntegerType() && !ValType->isPointerType() &&
416      !ValType->isBlockPointerType()) {
417    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr)
418      << FirstArg->getType() << FirstArg->getSourceRange();
419    return ExprError();
420  }
421
422  // The majority of builtins return a value, but a few have special return
423  // types, so allow them to override appropriately below.
424  QualType ResultType = ValType;
425
426  // We need to figure out which concrete builtin this maps onto.  For example,
427  // __sync_fetch_and_add with a 2 byte object turns into
428  // __sync_fetch_and_add_2.
429#define BUILTIN_ROW(x) \
430  { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
431    Builtin::BI##x##_8, Builtin::BI##x##_16 }
432
433  static const unsigned BuiltinIndices[][5] = {
434    BUILTIN_ROW(__sync_fetch_and_add),
435    BUILTIN_ROW(__sync_fetch_and_sub),
436    BUILTIN_ROW(__sync_fetch_and_or),
437    BUILTIN_ROW(__sync_fetch_and_and),
438    BUILTIN_ROW(__sync_fetch_and_xor),
439
440    BUILTIN_ROW(__sync_add_and_fetch),
441    BUILTIN_ROW(__sync_sub_and_fetch),
442    BUILTIN_ROW(__sync_and_and_fetch),
443    BUILTIN_ROW(__sync_or_and_fetch),
444    BUILTIN_ROW(__sync_xor_and_fetch),
445
446    BUILTIN_ROW(__sync_val_compare_and_swap),
447    BUILTIN_ROW(__sync_bool_compare_and_swap),
448    BUILTIN_ROW(__sync_lock_test_and_set),
449    BUILTIN_ROW(__sync_lock_release)
450  };
451#undef BUILTIN_ROW
452
453  // Determine the index of the size.
454  unsigned SizeIndex;
455  switch (Context.getTypeSizeInChars(ValType).getQuantity()) {
456  case 1: SizeIndex = 0; break;
457  case 2: SizeIndex = 1; break;
458  case 4: SizeIndex = 2; break;
459  case 8: SizeIndex = 3; break;
460  case 16: SizeIndex = 4; break;
461  default:
462    Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
463      << FirstArg->getType() << FirstArg->getSourceRange();
464    return ExprError();
465  }
466
467  // Each of these builtins has one pointer argument, followed by some number of
468  // values (0, 1 or 2) followed by a potentially empty varags list of stuff
469  // that we ignore.  Find out which row of BuiltinIndices to read from as well
470  // as the number of fixed args.
471  unsigned BuiltinID = FDecl->getBuiltinID();
472  unsigned BuiltinIndex, NumFixed = 1;
473  switch (BuiltinID) {
474  default: assert(0 && "Unknown overloaded atomic builtin!");
475  case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break;
476  case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break;
477  case Builtin::BI__sync_fetch_and_or:  BuiltinIndex = 2; break;
478  case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break;
479  case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break;
480
481  case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break;
482  case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break;
483  case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break;
484  case Builtin::BI__sync_or_and_fetch:  BuiltinIndex = 8; break;
485  case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break;
486
487  case Builtin::BI__sync_val_compare_and_swap:
488    BuiltinIndex = 10;
489    NumFixed = 2;
490    break;
491  case Builtin::BI__sync_bool_compare_and_swap:
492    BuiltinIndex = 11;
493    NumFixed = 2;
494    ResultType = Context.BoolTy;
495    break;
496  case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break;
497  case Builtin::BI__sync_lock_release:
498    BuiltinIndex = 13;
499    NumFixed = 0;
500    ResultType = Context.VoidTy;
501    break;
502  }
503
504  // Now that we know how many fixed arguments we expect, first check that we
505  // have at least that many.
506  if (TheCall->getNumArgs() < 1+NumFixed) {
507    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
508      << 0 << 1+NumFixed << TheCall->getNumArgs()
509      << TheCall->getCallee()->getSourceRange();
510    return ExprError();
511  }
512
513  // Get the decl for the concrete builtin from this, we can tell what the
514  // concrete integer type we should convert to is.
515  unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
516  const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
517  IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName);
518  FunctionDecl *NewBuiltinDecl =
519    cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID,
520                                           TUScope, false, DRE->getLocStart()));
521
522  // The first argument is by definition correct, we use it's type as the type
523  // of the entire operation. Walk the remaining arguments promoting them to
524  // the deduced value type.
525  for (unsigned i = 0; i != NumFixed; ++i) {
526    Expr *Arg = TheCall->getArg(i+1);
527
528    // If the argument is an implicit cast, then there was a promotion due to
529    // "...", just remove it now.
530    if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) {
531      Arg = ICE->getSubExpr();
532      ICE->setSubExpr(0);
533      TheCall->setArg(i+1, Arg);
534    }
535
536    // GCC does an implicit conversion to the pointer or integer ValType.  This
537    // can fail in some cases (1i -> int**), check for this error case now.
538    CastExpr::CastKind Kind = CastExpr::CK_Unknown;
539    CXXBaseSpecifierArray BasePath;
540    if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind, BasePath))
541      return ExprError();
542
543    // Okay, we have something that *can* be converted to the right type.  Check
544    // to see if there is a potentially weird extension going on here.  This can
545    // happen when you do an atomic operation on something like an char* and
546    // pass in 42.  The 42 gets converted to char.  This is even more strange
547    // for things like 45.123 -> char, etc.
548    // FIXME: Do this check.
549    ImpCastExprToType(Arg, ValType, Kind);
550    TheCall->setArg(i+1, Arg);
551  }
552
553  // Switch the DeclRefExpr to refer to the new decl.
554  DRE->setDecl(NewBuiltinDecl);
555  DRE->setType(NewBuiltinDecl->getType());
556
557  // Set the callee in the CallExpr.
558  // FIXME: This leaks the original parens and implicit casts.
559  Expr *PromotedCall = DRE;
560  UsualUnaryConversions(PromotedCall);
561  TheCall->setCallee(PromotedCall);
562
563  // Change the result type of the call to match the original value type. This
564  // is arbitrary, but the codegen for these builtins ins design to handle it
565  // gracefully.
566  TheCall->setType(ResultType);
567
568  return move(TheCallResult);
569}
570
571
572/// CheckObjCString - Checks that the argument to the builtin
573/// CFString constructor is correct
574/// FIXME: GCC currently emits the following warning:
575/// "warning: input conversion stopped due to an input byte that does not
576///           belong to the input codeset UTF-8"
577/// Note: It might also make sense to do the UTF-16 conversion here (would
578/// simplify the backend).
579bool Sema::CheckObjCString(Expr *Arg) {
580  Arg = Arg->IgnoreParenCasts();
581  StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
582
583  if (!Literal || Literal->isWide()) {
584    Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
585      << Arg->getSourceRange();
586    return true;
587  }
588
589  const char *Data = Literal->getStrData();
590  unsigned Length = Literal->getByteLength();
591
592  for (unsigned i = 0; i < Length; ++i) {
593    if (!Data[i]) {
594      Diag(getLocationOfStringLiteralByte(Literal, i),
595           diag::warn_cfstring_literal_contains_nul_character)
596        << Arg->getSourceRange();
597      break;
598    }
599  }
600
601  return false;
602}
603
604/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
605/// Emit an error and return true on failure, return false on success.
606bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
607  Expr *Fn = TheCall->getCallee();
608  if (TheCall->getNumArgs() > 2) {
609    Diag(TheCall->getArg(2)->getLocStart(),
610         diag::err_typecheck_call_too_many_args)
611      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
612      << Fn->getSourceRange()
613      << SourceRange(TheCall->getArg(2)->getLocStart(),
614                     (*(TheCall->arg_end()-1))->getLocEnd());
615    return true;
616  }
617
618  if (TheCall->getNumArgs() < 2) {
619    return Diag(TheCall->getLocEnd(),
620      diag::err_typecheck_call_too_few_args_at_least)
621      << 0 /*function call*/ << 2 << TheCall->getNumArgs();
622  }
623
624  // Determine whether the current function is variadic or not.
625  BlockScopeInfo *CurBlock = getCurBlock();
626  bool isVariadic;
627  if (CurBlock)
628    isVariadic = CurBlock->TheDecl->isVariadic();
629  else if (FunctionDecl *FD = getCurFunctionDecl())
630    isVariadic = FD->isVariadic();
631  else
632    isVariadic = getCurMethodDecl()->isVariadic();
633
634  if (!isVariadic) {
635    Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
636    return true;
637  }
638
639  // Verify that the second argument to the builtin is the last argument of the
640  // current function or method.
641  bool SecondArgIsLastNamedArgument = false;
642  const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
643
644  if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
645    if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
646      // FIXME: This isn't correct for methods (results in bogus warning).
647      // Get the last formal in the current function.
648      const ParmVarDecl *LastArg;
649      if (CurBlock)
650        LastArg = *(CurBlock->TheDecl->param_end()-1);
651      else if (FunctionDecl *FD = getCurFunctionDecl())
652        LastArg = *(FD->param_end()-1);
653      else
654        LastArg = *(getCurMethodDecl()->param_end()-1);
655      SecondArgIsLastNamedArgument = PV == LastArg;
656    }
657  }
658
659  if (!SecondArgIsLastNamedArgument)
660    Diag(TheCall->getArg(1)->getLocStart(),
661         diag::warn_second_parameter_of_va_start_not_last_named_argument);
662  return false;
663}
664
665/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
666/// friends.  This is declared to take (...), so we have to check everything.
667bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
668  if (TheCall->getNumArgs() < 2)
669    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
670      << 0 << 2 << TheCall->getNumArgs()/*function call*/;
671  if (TheCall->getNumArgs() > 2)
672    return Diag(TheCall->getArg(2)->getLocStart(),
673                diag::err_typecheck_call_too_many_args)
674      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
675      << SourceRange(TheCall->getArg(2)->getLocStart(),
676                     (*(TheCall->arg_end()-1))->getLocEnd());
677
678  Expr *OrigArg0 = TheCall->getArg(0);
679  Expr *OrigArg1 = TheCall->getArg(1);
680
681  // Do standard promotions between the two arguments, returning their common
682  // type.
683  QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
684
685  // Make sure any conversions are pushed back into the call; this is
686  // type safe since unordered compare builtins are declared as "_Bool
687  // foo(...)".
688  TheCall->setArg(0, OrigArg0);
689  TheCall->setArg(1, OrigArg1);
690
691  if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent())
692    return false;
693
694  // If the common type isn't a real floating type, then the arguments were
695  // invalid for this operation.
696  if (!Res->isRealFloatingType())
697    return Diag(OrigArg0->getLocStart(),
698                diag::err_typecheck_call_invalid_ordered_compare)
699      << OrigArg0->getType() << OrigArg1->getType()
700      << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd());
701
702  return false;
703}
704
705/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like
706/// __builtin_isnan and friends.  This is declared to take (...), so we have
707/// to check everything. We expect the last argument to be a floating point
708/// value.
709bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) {
710  if (TheCall->getNumArgs() < NumArgs)
711    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
712      << 0 << NumArgs << TheCall->getNumArgs()/*function call*/;
713  if (TheCall->getNumArgs() > NumArgs)
714    return Diag(TheCall->getArg(NumArgs)->getLocStart(),
715                diag::err_typecheck_call_too_many_args)
716      << 0 /*function call*/ << NumArgs << TheCall->getNumArgs()
717      << SourceRange(TheCall->getArg(NumArgs)->getLocStart(),
718                     (*(TheCall->arg_end()-1))->getLocEnd());
719
720  Expr *OrigArg = TheCall->getArg(NumArgs-1);
721
722  if (OrigArg->isTypeDependent())
723    return false;
724
725  // This operation requires a non-_Complex floating-point number.
726  if (!OrigArg->getType()->isRealFloatingType())
727    return Diag(OrigArg->getLocStart(),
728                diag::err_typecheck_call_invalid_unary_fp)
729      << OrigArg->getType() << OrigArg->getSourceRange();
730
731  // If this is an implicit conversion from float -> double, remove it.
732  if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) {
733    Expr *CastArg = Cast->getSubExpr();
734    if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) {
735      assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) &&
736             "promotion from float to double is the only expected cast here");
737      Cast->setSubExpr(0);
738      TheCall->setArg(NumArgs-1, CastArg);
739      OrigArg = CastArg;
740    }
741  }
742
743  return false;
744}
745
746/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
747// This is declared to take (...), so we have to check everything.
748Action::OwningExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
749  if (TheCall->getNumArgs() < 2)
750    return ExprError(Diag(TheCall->getLocEnd(),
751                          diag::err_typecheck_call_too_few_args_at_least)
752      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
753      << TheCall->getSourceRange());
754
755  // Determine which of the following types of shufflevector we're checking:
756  // 1) unary, vector mask: (lhs, mask)
757  // 2) binary, vector mask: (lhs, rhs, mask)
758  // 3) binary, scalar mask: (lhs, rhs, index, ..., index)
759  QualType resType = TheCall->getArg(0)->getType();
760  unsigned numElements = 0;
761
762  if (!TheCall->getArg(0)->isTypeDependent() &&
763      !TheCall->getArg(1)->isTypeDependent()) {
764    QualType LHSType = TheCall->getArg(0)->getType();
765    QualType RHSType = TheCall->getArg(1)->getType();
766
767    if (!LHSType->isVectorType() || !RHSType->isVectorType()) {
768      Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
769        << SourceRange(TheCall->getArg(0)->getLocStart(),
770                       TheCall->getArg(1)->getLocEnd());
771      return ExprError();
772    }
773
774    numElements = LHSType->getAs<VectorType>()->getNumElements();
775    unsigned numResElements = TheCall->getNumArgs() - 2;
776
777    // Check to see if we have a call with 2 vector arguments, the unary shuffle
778    // with mask.  If so, verify that RHS is an integer vector type with the
779    // same number of elts as lhs.
780    if (TheCall->getNumArgs() == 2) {
781      if (!RHSType->hasIntegerRepresentation() ||
782          RHSType->getAs<VectorType>()->getNumElements() != numElements)
783        Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
784          << SourceRange(TheCall->getArg(1)->getLocStart(),
785                         TheCall->getArg(1)->getLocEnd());
786      numResElements = numElements;
787    }
788    else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) {
789      Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
790        << SourceRange(TheCall->getArg(0)->getLocStart(),
791                       TheCall->getArg(1)->getLocEnd());
792      return ExprError();
793    } else if (numElements != numResElements) {
794      QualType eltType = LHSType->getAs<VectorType>()->getElementType();
795      resType = Context.getVectorType(eltType, numResElements,
796                                      VectorType::NotAltiVec);
797    }
798  }
799
800  for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
801    if (TheCall->getArg(i)->isTypeDependent() ||
802        TheCall->getArg(i)->isValueDependent())
803      continue;
804
805    llvm::APSInt Result(32);
806    if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
807      return ExprError(Diag(TheCall->getLocStart(),
808                  diag::err_shufflevector_nonconstant_argument)
809                << TheCall->getArg(i)->getSourceRange());
810
811    if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
812      return ExprError(Diag(TheCall->getLocStart(),
813                  diag::err_shufflevector_argument_too_large)
814               << TheCall->getArg(i)->getSourceRange());
815  }
816
817  llvm::SmallVector<Expr*, 32> exprs;
818
819  for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
820    exprs.push_back(TheCall->getArg(i));
821    TheCall->setArg(i, 0);
822  }
823
824  return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(),
825                                            exprs.size(), resType,
826                                            TheCall->getCallee()->getLocStart(),
827                                            TheCall->getRParenLoc()));
828}
829
830/// SemaBuiltinPrefetch - Handle __builtin_prefetch.
831// This is declared to take (const void*, ...) and can take two
832// optional constant int args.
833bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
834  unsigned NumArgs = TheCall->getNumArgs();
835
836  if (NumArgs > 3)
837    return Diag(TheCall->getLocEnd(),
838             diag::err_typecheck_call_too_many_args_at_most)
839             << 0 /*function call*/ << 3 << NumArgs
840             << TheCall->getSourceRange();
841
842  // Argument 0 is checked for us and the remaining arguments must be
843  // constant integers.
844  for (unsigned i = 1; i != NumArgs; ++i) {
845    Expr *Arg = TheCall->getArg(i);
846
847    llvm::APSInt Result;
848    if (SemaBuiltinConstantArg(TheCall, i, Result))
849      return true;
850
851    // FIXME: gcc issues a warning and rewrites these to 0. These
852    // seems especially odd for the third argument since the default
853    // is 3.
854    if (i == 1) {
855      if (Result.getLimitedValue() > 1)
856        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
857             << "0" << "1" << Arg->getSourceRange();
858    } else {
859      if (Result.getLimitedValue() > 3)
860        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
861            << "0" << "3" << Arg->getSourceRange();
862    }
863  }
864
865  return false;
866}
867
868/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr
869/// TheCall is a constant expression.
870bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,
871                                  llvm::APSInt &Result) {
872  Expr *Arg = TheCall->getArg(ArgNum);
873  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
874  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
875
876  if (Arg->isTypeDependent() || Arg->isValueDependent()) return false;
877
878  if (!Arg->isIntegerConstantExpr(Result, Context))
879    return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type)
880                << FDecl->getDeclName() <<  Arg->getSourceRange();
881
882  return false;
883}
884
885/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
886/// int type). This simply type checks that type is one of the defined
887/// constants (0-3).
888// For compatability check 0-3, llvm only handles 0 and 2.
889bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
890  llvm::APSInt Result;
891
892  // Check constant-ness first.
893  if (SemaBuiltinConstantArg(TheCall, 1, Result))
894    return true;
895
896  Expr *Arg = TheCall->getArg(1);
897  if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
898    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
899             << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
900  }
901
902  return false;
903}
904
905/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
906/// This checks that val is a constant 1.
907bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
908  Expr *Arg = TheCall->getArg(1);
909  llvm::APSInt Result;
910
911  // TODO: This is less than ideal. Overload this to take a value.
912  if (SemaBuiltinConstantArg(TheCall, 1, Result))
913    return true;
914
915  if (Result != 1)
916    return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
917             << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
918
919  return false;
920}
921
922// Handle i > 1 ? "x" : "y", recursivelly
923bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall,
924                                  bool HasVAListArg,
925                                  unsigned format_idx, unsigned firstDataArg,
926                                  bool isPrintf) {
927
928  if (E->isTypeDependent() || E->isValueDependent())
929    return false;
930
931  switch (E->getStmtClass()) {
932  case Stmt::ConditionalOperatorClass: {
933    const ConditionalOperator *C = cast<ConditionalOperator>(E);
934    return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg,
935                                  format_idx, firstDataArg, isPrintf)
936        && SemaCheckStringLiteral(C->getRHS(), TheCall, HasVAListArg,
937                                  format_idx, firstDataArg, isPrintf);
938  }
939
940  case Stmt::ImplicitCastExprClass: {
941    const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E);
942    return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg,
943                                  format_idx, firstDataArg, isPrintf);
944  }
945
946  case Stmt::ParenExprClass: {
947    const ParenExpr *Expr = cast<ParenExpr>(E);
948    return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg,
949                                  format_idx, firstDataArg, isPrintf);
950  }
951
952  case Stmt::DeclRefExprClass: {
953    const DeclRefExpr *DR = cast<DeclRefExpr>(E);
954
955    // As an exception, do not flag errors for variables binding to
956    // const string literals.
957    if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
958      bool isConstant = false;
959      QualType T = DR->getType();
960
961      if (const ArrayType *AT = Context.getAsArrayType(T)) {
962        isConstant = AT->getElementType().isConstant(Context);
963      } else if (const PointerType *PT = T->getAs<PointerType>()) {
964        isConstant = T.isConstant(Context) &&
965                     PT->getPointeeType().isConstant(Context);
966      }
967
968      if (isConstant) {
969        if (const Expr *Init = VD->getAnyInitializer())
970          return SemaCheckStringLiteral(Init, TheCall,
971                                        HasVAListArg, format_idx, firstDataArg,
972                                        isPrintf);
973      }
974
975      // For vprintf* functions (i.e., HasVAListArg==true), we add a
976      // special check to see if the format string is a function parameter
977      // of the function calling the printf function.  If the function
978      // has an attribute indicating it is a printf-like function, then we
979      // should suppress warnings concerning non-literals being used in a call
980      // to a vprintf function.  For example:
981      //
982      // void
983      // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
984      //      va_list ap;
985      //      va_start(ap, fmt);
986      //      vprintf(fmt, ap);  // Do NOT emit a warning about "fmt".
987      //      ...
988      //
989      //
990      //  FIXME: We don't have full attribute support yet, so just check to see
991      //    if the argument is a DeclRefExpr that references a parameter.  We'll
992      //    add proper support for checking the attribute later.
993      if (HasVAListArg)
994        if (isa<ParmVarDecl>(VD))
995          return true;
996    }
997
998    return false;
999  }
1000
1001  case Stmt::CallExprClass: {
1002    const CallExpr *CE = cast<CallExpr>(E);
1003    if (const ImplicitCastExpr *ICE
1004          = dyn_cast<ImplicitCastExpr>(CE->getCallee())) {
1005      if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) {
1006        if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) {
1007          if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) {
1008            unsigned ArgIndex = FA->getFormatIdx();
1009            const Expr *Arg = CE->getArg(ArgIndex - 1);
1010
1011            return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg,
1012                                          format_idx, firstDataArg, isPrintf);
1013          }
1014        }
1015      }
1016    }
1017
1018    return false;
1019  }
1020  case Stmt::ObjCStringLiteralClass:
1021  case Stmt::StringLiteralClass: {
1022    const StringLiteral *StrE = NULL;
1023
1024    if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
1025      StrE = ObjCFExpr->getString();
1026    else
1027      StrE = cast<StringLiteral>(E);
1028
1029    if (StrE) {
1030      CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx,
1031                        firstDataArg, isPrintf);
1032      return true;
1033    }
1034
1035    return false;
1036  }
1037
1038  default:
1039    return false;
1040  }
1041}
1042
1043void
1044Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
1045                            const CallExpr *TheCall) {
1046  for (NonNullAttr::iterator i = NonNull->begin(), e = NonNull->end();
1047       i != e; ++i) {
1048    const Expr *ArgExpr = TheCall->getArg(*i);
1049    if (ArgExpr->isNullPointerConstant(Context,
1050                                       Expr::NPC_ValueDependentIsNotNull))
1051      Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg)
1052        << ArgExpr->getSourceRange();
1053  }
1054}
1055
1056/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar
1057/// functions) for correct use of format strings.
1058void
1059Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg,
1060                                unsigned format_idx, unsigned firstDataArg,
1061                                bool isPrintf) {
1062
1063  const Expr *Fn = TheCall->getCallee();
1064
1065  // The way the format attribute works in GCC, the implicit this argument
1066  // of member functions is counted. However, it doesn't appear in our own
1067  // lists, so decrement format_idx in that case.
1068  if (isa<CXXMemberCallExpr>(TheCall)) {
1069    // 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  // CHECK: printf/scanf-like function is called with no format string.
1078  if (format_idx >= TheCall->getNumArgs()) {
1079    Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string)
1080      << Fn->getSourceRange();
1081    return;
1082  }
1083
1084  const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts();
1085
1086  // CHECK: format string is not a string literal.
1087  //
1088  // Dynamically generated format strings are difficult to
1089  // automatically vet at compile time.  Requiring that format strings
1090  // are string literals: (1) permits the checking of format strings by
1091  // the compiler and thereby (2) can practically remove the source of
1092  // many format string exploits.
1093
1094  // Format string can be either ObjC string (e.g. @"%d") or
1095  // C string (e.g. "%d")
1096  // ObjC string uses the same format specifiers as C string, so we can use
1097  // the same format string checking logic for both ObjC and C strings.
1098  if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx,
1099                             firstDataArg, isPrintf))
1100    return;  // Literal format string found, check done!
1101
1102  // If there are no arguments specified, warn with -Wformat-security, otherwise
1103  // warn only with -Wformat-nonliteral.
1104  if (TheCall->getNumArgs() == format_idx+1)
1105    Diag(TheCall->getArg(format_idx)->getLocStart(),
1106         diag::warn_format_nonliteral_noargs)
1107      << OrigFormatExpr->getSourceRange();
1108  else
1109    Diag(TheCall->getArg(format_idx)->getLocStart(),
1110         diag::warn_format_nonliteral)
1111           << OrigFormatExpr->getSourceRange();
1112}
1113
1114namespace {
1115class CheckFormatHandler : public analyze_format_string::FormatStringHandler {
1116protected:
1117  Sema &S;
1118  const StringLiteral *FExpr;
1119  const Expr *OrigFormatExpr;
1120  const unsigned FirstDataArg;
1121  const unsigned NumDataArgs;
1122  const bool IsObjCLiteral;
1123  const char *Beg; // Start of format string.
1124  const bool HasVAListArg;
1125  const CallExpr *TheCall;
1126  unsigned FormatIdx;
1127  llvm::BitVector CoveredArgs;
1128  bool usesPositionalArgs;
1129  bool atFirstArg;
1130public:
1131  CheckFormatHandler(Sema &s, const StringLiteral *fexpr,
1132                     const Expr *origFormatExpr, unsigned firstDataArg,
1133                     unsigned numDataArgs, bool isObjCLiteral,
1134                     const char *beg, bool hasVAListArg,
1135                     const CallExpr *theCall, unsigned formatIdx)
1136    : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr),
1137      FirstDataArg(firstDataArg),
1138      NumDataArgs(numDataArgs),
1139      IsObjCLiteral(isObjCLiteral), Beg(beg),
1140      HasVAListArg(hasVAListArg),
1141      TheCall(theCall), FormatIdx(formatIdx),
1142      usesPositionalArgs(false), atFirstArg(true) {
1143        CoveredArgs.resize(numDataArgs);
1144        CoveredArgs.reset();
1145      }
1146
1147  void DoneProcessing();
1148
1149  void HandleIncompleteSpecifier(const char *startSpecifier,
1150                                 unsigned specifierLen);
1151
1152  virtual void HandleInvalidPosition(const char *startSpecifier,
1153                                     unsigned specifierLen,
1154                                     analyze_format_string::PositionContext p);
1155
1156  virtual void HandleZeroPosition(const char *startPos, unsigned posLen);
1157
1158  void HandleNullChar(const char *nullCharacter);
1159
1160protected:
1161  bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc,
1162                                        const char *startSpec,
1163                                        unsigned specifierLen,
1164                                        const char *csStart, unsigned csLen);
1165
1166  SourceRange getFormatStringRange();
1167  CharSourceRange getSpecifierRange(const char *startSpecifier,
1168                                    unsigned specifierLen);
1169  SourceLocation getLocationOfByte(const char *x);
1170
1171  const Expr *getDataArg(unsigned i) const;
1172
1173  bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS,
1174                    const analyze_format_string::ConversionSpecifier &CS,
1175                    const char *startSpecifier, unsigned specifierLen,
1176                    unsigned argIndex);
1177
1178  void CheckArgType(const analyze_format_string::FormatSpecifier &FS,
1179                    const analyze_format_string::ConversionSpecifier &CS,
1180                    const char *startSpecifier, unsigned specifierLen,
1181                    unsigned argIndex);
1182};
1183}
1184
1185SourceRange CheckFormatHandler::getFormatStringRange() {
1186  return OrigFormatExpr->getSourceRange();
1187}
1188
1189CharSourceRange CheckFormatHandler::
1190getSpecifierRange(const char *startSpecifier, unsigned specifierLen) {
1191  SourceLocation Start = getLocationOfByte(startSpecifier);
1192  SourceLocation End   = getLocationOfByte(startSpecifier + specifierLen - 1);
1193
1194  // Advance the end SourceLocation by one due to half-open ranges.
1195  End = End.getFileLocWithOffset(1);
1196
1197  return CharSourceRange::getCharRange(Start, End);
1198}
1199
1200SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) {
1201  return S.getLocationOfStringLiteralByte(FExpr, x - Beg);
1202}
1203
1204void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier,
1205                                                   unsigned specifierLen){
1206  SourceLocation Loc = getLocationOfByte(startSpecifier);
1207  S.Diag(Loc, diag::warn_printf_incomplete_specifier)
1208    << getSpecifierRange(startSpecifier, specifierLen);
1209}
1210
1211void
1212CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen,
1213                                     analyze_format_string::PositionContext p) {
1214  SourceLocation Loc = getLocationOfByte(startPos);
1215  S.Diag(Loc, diag::warn_format_invalid_positional_specifier)
1216    << (unsigned) p << getSpecifierRange(startPos, posLen);
1217}
1218
1219void CheckFormatHandler::HandleZeroPosition(const char *startPos,
1220                                            unsigned posLen) {
1221  SourceLocation Loc = getLocationOfByte(startPos);
1222  S.Diag(Loc, diag::warn_format_zero_positional_specifier)
1223    << getSpecifierRange(startPos, posLen);
1224}
1225
1226void CheckFormatHandler::HandleNullChar(const char *nullCharacter) {
1227  // The presence of a null character is likely an error.
1228  S.Diag(getLocationOfByte(nullCharacter),
1229         diag::warn_printf_format_string_contains_null_char)
1230    << getFormatStringRange();
1231}
1232
1233const Expr *CheckFormatHandler::getDataArg(unsigned i) const {
1234  return TheCall->getArg(FirstDataArg + i);
1235}
1236
1237void CheckFormatHandler::DoneProcessing() {
1238    // Does the number of data arguments exceed the number of
1239    // format conversions in the format string?
1240  if (!HasVAListArg) {
1241      // Find any arguments that weren't covered.
1242    CoveredArgs.flip();
1243    signed notCoveredArg = CoveredArgs.find_first();
1244    if (notCoveredArg >= 0) {
1245      assert((unsigned)notCoveredArg < NumDataArgs);
1246      S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(),
1247             diag::warn_printf_data_arg_not_used)
1248      << getFormatStringRange();
1249    }
1250  }
1251}
1252
1253bool
1254CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex,
1255                                                     SourceLocation Loc,
1256                                                     const char *startSpec,
1257                                                     unsigned specifierLen,
1258                                                     const char *csStart,
1259                                                     unsigned csLen) {
1260
1261  bool keepGoing = true;
1262  if (argIndex < NumDataArgs) {
1263    // Consider the argument coverered, even though the specifier doesn't
1264    // make sense.
1265    CoveredArgs.set(argIndex);
1266  }
1267  else {
1268    // If argIndex exceeds the number of data arguments we
1269    // don't issue a warning because that is just a cascade of warnings (and
1270    // they may have intended '%%' anyway). We don't want to continue processing
1271    // the format string after this point, however, as we will like just get
1272    // gibberish when trying to match arguments.
1273    keepGoing = false;
1274  }
1275
1276  S.Diag(Loc, diag::warn_format_invalid_conversion)
1277    << llvm::StringRef(csStart, csLen)
1278    << getSpecifierRange(startSpec, specifierLen);
1279
1280  return keepGoing;
1281}
1282
1283bool
1284CheckFormatHandler::CheckNumArgs(
1285  const analyze_format_string::FormatSpecifier &FS,
1286  const analyze_format_string::ConversionSpecifier &CS,
1287  const char *startSpecifier, unsigned specifierLen, unsigned argIndex) {
1288
1289  if (argIndex >= NumDataArgs) {
1290    if (FS.usesPositionalArg())  {
1291      S.Diag(getLocationOfByte(CS.getStart()),
1292             diag::warn_printf_positional_arg_exceeds_data_args)
1293      << (argIndex+1) << NumDataArgs
1294      << getSpecifierRange(startSpecifier, specifierLen);
1295    }
1296    else {
1297      S.Diag(getLocationOfByte(CS.getStart()),
1298             diag::warn_printf_insufficient_data_args)
1299      << getSpecifierRange(startSpecifier, specifierLen);
1300    }
1301
1302    return false;
1303  }
1304  return true;
1305}
1306
1307void CheckFormatHandler::CheckArgType(
1308  const analyze_format_string::FormatSpecifier &FS,
1309  const analyze_format_string::ConversionSpecifier &CS,
1310  const char *startSpecifier, unsigned specifierLen, unsigned argIndex) {
1311
1312  const Expr *Ex = getDataArg(argIndex);
1313  const analyze_format_string::ArgTypeResult &ATR = FS.getArgType(S.Context);
1314
1315  if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) {
1316    // Check if we didn't match because of an implicit cast from a 'char'
1317    // or 'short' to an 'int'.  This is done because scanf/printf are varargs
1318    // functions.
1319    if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex))
1320      if (ICE->getType() == S.Context.IntTy)
1321        if (ATR.matchesType(S.Context, ICE->getSubExpr()->getType()))
1322          return;
1323
1324    if (const analyze_printf::PrintfSpecifier *PFS =
1325        dyn_cast<analyze_printf::PrintfSpecifier>(&FS)) {
1326      // We may be able to offer a FixItHint if it is a supported type.
1327      analyze_printf::PrintfSpecifier fixedFS(*PFS);
1328      if (fixedFS.fixType(Ex->getType())) {
1329        // Get the fix string from the fixed format specifier
1330        llvm::SmallString<128> buf;
1331        llvm::raw_svector_ostream os(buf);
1332        fixedFS.toString(os);
1333
1334        S.Diag(getLocationOfByte(CS.getStart()),
1335               diag::warn_printf_conversion_argument_type_mismatch)
1336          << ATR.getRepresentativeType(S.Context) << Ex->getType()
1337          << getSpecifierRange(startSpecifier, specifierLen)
1338        << Ex->getSourceRange()
1339        << FixItHint::CreateReplacement(
1340                getSpecifierRange(startSpecifier, specifierLen), os.str());
1341      }
1342      else {
1343        S.Diag(getLocationOfByte(CS.getStart()),
1344               diag::warn_printf_conversion_argument_type_mismatch)
1345          << ATR.getRepresentativeType(S.Context) << Ex->getType()
1346          << getSpecifierRange(startSpecifier, specifierLen)
1347          << Ex->getSourceRange();
1348      }
1349    }
1350  }
1351}
1352
1353//===--- CHECK: Printf format string checking ------------------------------===//
1354
1355namespace {
1356class CheckPrintfHandler : public CheckFormatHandler {
1357public:
1358  CheckPrintfHandler(Sema &s, const StringLiteral *fexpr,
1359                     const Expr *origFormatExpr, unsigned firstDataArg,
1360                     unsigned numDataArgs, bool isObjCLiteral,
1361                     const char *beg, bool hasVAListArg,
1362                     const CallExpr *theCall, unsigned formatIdx)
1363  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1364                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1365                       theCall, formatIdx) {}
1366
1367
1368  bool HandleInvalidPrintfConversionSpecifier(
1369                                      const analyze_printf::PrintfSpecifier &FS,
1370                                      const char *startSpecifier,
1371                                      unsigned specifierLen);
1372
1373  bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS,
1374                             const char *startSpecifier,
1375                             unsigned specifierLen);
1376
1377  bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k,
1378                    const char *startSpecifier, unsigned specifierLen);
1379  void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS,
1380                           const analyze_printf::OptionalAmount &Amt,
1381                           unsigned type,
1382                           const char *startSpecifier, unsigned specifierLen);
1383  void HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1384                  const analyze_printf::OptionalFlag &flag,
1385                  const char *startSpecifier, unsigned specifierLen);
1386  void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS,
1387                         const analyze_printf::OptionalFlag &ignoredFlag,
1388                         const analyze_printf::OptionalFlag &flag,
1389                         const char *startSpecifier, unsigned specifierLen);
1390};
1391}
1392
1393bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier(
1394                                      const analyze_printf::PrintfSpecifier &FS,
1395                                      const char *startSpecifier,
1396                                      unsigned specifierLen) {
1397  const analyze_printf::PrintfConversionSpecifier &CS =
1398    FS.getConversionSpecifier();
1399
1400  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1401                                          getLocationOfByte(CS.getStart()),
1402                                          startSpecifier, specifierLen,
1403                                          CS.getStart(), CS.getLength());
1404}
1405
1406bool CheckPrintfHandler::HandleAmount(
1407                               const analyze_format_string::OptionalAmount &Amt,
1408                               unsigned k, const char *startSpecifier,
1409                               unsigned specifierLen) {
1410
1411  if (Amt.hasDataArgument()) {
1412    if (!HasVAListArg) {
1413      unsigned argIndex = Amt.getArgIndex();
1414      if (argIndex >= NumDataArgs) {
1415        S.Diag(getLocationOfByte(Amt.getStart()),
1416               diag::warn_printf_asterisk_missing_arg)
1417          << k << getSpecifierRange(startSpecifier, specifierLen);
1418        // Don't do any more checking.  We will just emit
1419        // spurious errors.
1420        return false;
1421      }
1422
1423      // Type check the data argument.  It should be an 'int'.
1424      // Although not in conformance with C99, we also allow the argument to be
1425      // an 'unsigned int' as that is a reasonably safe case.  GCC also
1426      // doesn't emit a warning for that case.
1427      CoveredArgs.set(argIndex);
1428      const Expr *Arg = getDataArg(argIndex);
1429      QualType T = Arg->getType();
1430
1431      const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context);
1432      assert(ATR.isValid());
1433
1434      if (!ATR.matchesType(S.Context, T)) {
1435        S.Diag(getLocationOfByte(Amt.getStart()),
1436               diag::warn_printf_asterisk_wrong_type)
1437          << k
1438          << ATR.getRepresentativeType(S.Context) << T
1439          << getSpecifierRange(startSpecifier, specifierLen)
1440          << Arg->getSourceRange();
1441        // Don't do any more checking.  We will just emit
1442        // spurious errors.
1443        return false;
1444      }
1445    }
1446  }
1447  return true;
1448}
1449
1450void CheckPrintfHandler::HandleInvalidAmount(
1451                                      const analyze_printf::PrintfSpecifier &FS,
1452                                      const analyze_printf::OptionalAmount &Amt,
1453                                      unsigned type,
1454                                      const char *startSpecifier,
1455                                      unsigned specifierLen) {
1456  const analyze_printf::PrintfConversionSpecifier &CS =
1457    FS.getConversionSpecifier();
1458  switch (Amt.getHowSpecified()) {
1459  case analyze_printf::OptionalAmount::Constant:
1460    S.Diag(getLocationOfByte(Amt.getStart()),
1461        diag::warn_printf_nonsensical_optional_amount)
1462      << type
1463      << CS.toString()
1464      << getSpecifierRange(startSpecifier, specifierLen)
1465      << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(),
1466          Amt.getConstantLength()));
1467    break;
1468
1469  default:
1470    S.Diag(getLocationOfByte(Amt.getStart()),
1471        diag::warn_printf_nonsensical_optional_amount)
1472      << type
1473      << CS.toString()
1474      << getSpecifierRange(startSpecifier, specifierLen);
1475    break;
1476  }
1477}
1478
1479void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1480                                    const analyze_printf::OptionalFlag &flag,
1481                                    const char *startSpecifier,
1482                                    unsigned specifierLen) {
1483  // Warn about pointless flag with a fixit removal.
1484  const analyze_printf::PrintfConversionSpecifier &CS =
1485    FS.getConversionSpecifier();
1486  S.Diag(getLocationOfByte(flag.getPosition()),
1487      diag::warn_printf_nonsensical_flag)
1488    << flag.toString() << CS.toString()
1489    << getSpecifierRange(startSpecifier, specifierLen)
1490    << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1));
1491}
1492
1493void CheckPrintfHandler::HandleIgnoredFlag(
1494                                const analyze_printf::PrintfSpecifier &FS,
1495                                const analyze_printf::OptionalFlag &ignoredFlag,
1496                                const analyze_printf::OptionalFlag &flag,
1497                                const char *startSpecifier,
1498                                unsigned specifierLen) {
1499  // Warn about ignored flag with a fixit removal.
1500  S.Diag(getLocationOfByte(ignoredFlag.getPosition()),
1501      diag::warn_printf_ignored_flag)
1502    << ignoredFlag.toString() << flag.toString()
1503    << getSpecifierRange(startSpecifier, specifierLen)
1504    << FixItHint::CreateRemoval(getSpecifierRange(
1505        ignoredFlag.getPosition(), 1));
1506}
1507
1508bool
1509CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
1510                                            &FS,
1511                                          const char *startSpecifier,
1512                                          unsigned specifierLen) {
1513
1514  using namespace analyze_format_string;
1515  using namespace analyze_printf;
1516  const PrintfConversionSpecifier &CS = FS.getConversionSpecifier();
1517
1518  if (FS.consumesDataArgument()) {
1519    if (atFirstArg) {
1520        atFirstArg = false;
1521        usesPositionalArgs = FS.usesPositionalArg();
1522    }
1523    else if (usesPositionalArgs != FS.usesPositionalArg()) {
1524      // Cannot mix-and-match positional and non-positional arguments.
1525      S.Diag(getLocationOfByte(CS.getStart()),
1526             diag::warn_format_mix_positional_nonpositional_args)
1527        << getSpecifierRange(startSpecifier, specifierLen);
1528      return false;
1529    }
1530  }
1531
1532  // First check if the field width, precision, and conversion specifier
1533  // have matching data arguments.
1534  if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0,
1535                    startSpecifier, specifierLen)) {
1536    return false;
1537  }
1538
1539  if (!HandleAmount(FS.getPrecision(), /* precision */ 1,
1540                    startSpecifier, specifierLen)) {
1541    return false;
1542  }
1543
1544  if (!CS.consumesDataArgument()) {
1545    // FIXME: Technically specifying a precision or field width here
1546    // makes no sense.  Worth issuing a warning at some point.
1547    return true;
1548  }
1549
1550  // Consume the argument.
1551  unsigned argIndex = FS.getArgIndex();
1552  if (argIndex < NumDataArgs) {
1553    // The check to see if the argIndex is valid will come later.
1554    // We set the bit here because we may exit early from this
1555    // function if we encounter some other error.
1556    CoveredArgs.set(argIndex);
1557  }
1558
1559  // Check for using an Objective-C specific conversion specifier
1560  // in a non-ObjC literal.
1561  if (!IsObjCLiteral && CS.isObjCArg()) {
1562    return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier,
1563                                                  specifierLen);
1564  }
1565
1566  // Check for invalid use of field width
1567  if (!FS.hasValidFieldWidth()) {
1568    HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0,
1569        startSpecifier, specifierLen);
1570  }
1571
1572  // Check for invalid use of precision
1573  if (!FS.hasValidPrecision()) {
1574    HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1,
1575        startSpecifier, specifierLen);
1576  }
1577
1578  // Check each flag does not conflict with any other component.
1579  if (!FS.hasValidLeadingZeros())
1580    HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen);
1581  if (!FS.hasValidPlusPrefix())
1582    HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen);
1583  if (!FS.hasValidSpacePrefix())
1584    HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen);
1585  if (!FS.hasValidAlternativeForm())
1586    HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen);
1587  if (!FS.hasValidLeftJustified())
1588    HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen);
1589
1590  // Check that flags are not ignored by another flag
1591  if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+'
1592    HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(),
1593        startSpecifier, specifierLen);
1594  if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-'
1595    HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(),
1596            startSpecifier, specifierLen);
1597
1598  // Check the length modifier is valid with the given conversion specifier.
1599  const LengthModifier &LM = FS.getLengthModifier();
1600  if (!FS.hasValidLengthModifier())
1601    S.Diag(getLocationOfByte(LM.getStart()),
1602        diag::warn_format_nonsensical_length)
1603      << LM.toString() << CS.toString()
1604      << getSpecifierRange(startSpecifier, specifierLen)
1605      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1606          LM.getLength()));
1607
1608  // Are we using '%n'?
1609  if (CS.getKind() == ConversionSpecifier::nArg) {
1610    // Issue a warning about this being a possible security issue.
1611    S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back)
1612      << getSpecifierRange(startSpecifier, specifierLen);
1613    // Continue checking the other format specifiers.
1614    return true;
1615  }
1616
1617  // The remaining checks depend on the data arguments.
1618  if (HasVAListArg)
1619    return true;
1620
1621  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
1622    return false;
1623
1624  CheckArgType(FS, CS, startSpecifier, specifierLen, argIndex);
1625
1626  return true;
1627}
1628
1629//===--- CHECK: Scanf format string checking ------------------------------===//
1630
1631namespace {
1632class CheckScanfHandler : public CheckFormatHandler {
1633public:
1634  CheckScanfHandler(Sema &s, const StringLiteral *fexpr,
1635                    const Expr *origFormatExpr, unsigned firstDataArg,
1636                    unsigned numDataArgs, bool isObjCLiteral,
1637                    const char *beg, bool hasVAListArg,
1638                    const CallExpr *theCall, unsigned formatIdx)
1639  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1640                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1641                       theCall, formatIdx) {}
1642
1643  bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS,
1644                            const char *startSpecifier,
1645                            unsigned specifierLen);
1646
1647  bool HandleInvalidScanfConversionSpecifier(
1648          const analyze_scanf::ScanfSpecifier &FS,
1649          const char *startSpecifier,
1650          unsigned specifierLen);
1651
1652  void HandleIncompleteScanList(const char *start, const char *end);
1653};
1654}
1655
1656void CheckScanfHandler::HandleIncompleteScanList(const char *start,
1657                                                 const char *end) {
1658  S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete)
1659    << getSpecifierRange(start, end - start);
1660}
1661
1662bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier(
1663                                        const analyze_scanf::ScanfSpecifier &FS,
1664                                        const char *startSpecifier,
1665                                        unsigned specifierLen) {
1666
1667  const analyze_scanf::ScanfConversionSpecifier &CS =
1668    FS.getConversionSpecifier();
1669
1670  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1671                                          getLocationOfByte(CS.getStart()),
1672                                          startSpecifier, specifierLen,
1673                                          CS.getStart(), CS.getLength());
1674}
1675
1676bool CheckScanfHandler::HandleScanfSpecifier(
1677                                       const analyze_scanf::ScanfSpecifier &FS,
1678                                       const char *startSpecifier,
1679                                       unsigned specifierLen) {
1680
1681  using namespace analyze_scanf;
1682  using namespace analyze_format_string;
1683
1684  const ScanfConversionSpecifier &CS = FS.getConversionSpecifier();
1685
1686  // Handle case where '%' and '*' don't consume an argument.  These shouldn't
1687  // be used to decide if we are using positional arguments consistently.
1688  if (FS.consumesDataArgument()) {
1689    if (atFirstArg) {
1690      atFirstArg = false;
1691      usesPositionalArgs = FS.usesPositionalArg();
1692    }
1693    else if (usesPositionalArgs != FS.usesPositionalArg()) {
1694      // Cannot mix-and-match positional and non-positional arguments.
1695      S.Diag(getLocationOfByte(CS.getStart()),
1696             diag::warn_format_mix_positional_nonpositional_args)
1697        << getSpecifierRange(startSpecifier, specifierLen);
1698      return false;
1699    }
1700  }
1701
1702  // Check if the field with is non-zero.
1703  const OptionalAmount &Amt = FS.getFieldWidth();
1704  if (Amt.getHowSpecified() == OptionalAmount::Constant) {
1705    if (Amt.getConstantAmount() == 0) {
1706      const CharSourceRange &R = getSpecifierRange(Amt.getStart(),
1707                                                   Amt.getConstantLength());
1708      S.Diag(getLocationOfByte(Amt.getStart()),
1709             diag::warn_scanf_nonzero_width)
1710        << R << FixItHint::CreateRemoval(R);
1711    }
1712  }
1713
1714  if (!FS.consumesDataArgument()) {
1715    // FIXME: Technically specifying a precision or field width here
1716    // makes no sense.  Worth issuing a warning at some point.
1717    return true;
1718  }
1719
1720  // Consume the argument.
1721  unsigned argIndex = FS.getArgIndex();
1722  if (argIndex < NumDataArgs) {
1723      // The check to see if the argIndex is valid will come later.
1724      // We set the bit here because we may exit early from this
1725      // function if we encounter some other error.
1726    CoveredArgs.set(argIndex);
1727  }
1728
1729  // Check the length modifier is valid with the given conversion specifier.
1730  const LengthModifier &LM = FS.getLengthModifier();
1731  if (!FS.hasValidLengthModifier()) {
1732    S.Diag(getLocationOfByte(LM.getStart()),
1733           diag::warn_format_nonsensical_length)
1734      << LM.toString() << CS.toString()
1735      << getSpecifierRange(startSpecifier, specifierLen)
1736      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1737                                                    LM.getLength()));
1738  }
1739
1740  // The remaining checks depend on the data arguments.
1741  if (HasVAListArg)
1742    return true;
1743
1744  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
1745    return false;
1746
1747  // FIXME: Check that the argument type matches the format specifier.
1748
1749  return true;
1750}
1751
1752void Sema::CheckFormatString(const StringLiteral *FExpr,
1753                             const Expr *OrigFormatExpr,
1754                             const CallExpr *TheCall, bool HasVAListArg,
1755                             unsigned format_idx, unsigned firstDataArg,
1756                             bool isPrintf) {
1757
1758  // CHECK: is the format string a wide literal?
1759  if (FExpr->isWide()) {
1760    Diag(FExpr->getLocStart(),
1761         diag::warn_format_string_is_wide_literal)
1762    << OrigFormatExpr->getSourceRange();
1763    return;
1764  }
1765
1766  // Str - The format string.  NOTE: this is NOT null-terminated!
1767  const char *Str = FExpr->getStrData();
1768
1769  // CHECK: empty format string?
1770  unsigned StrLen = FExpr->getByteLength();
1771
1772  if (StrLen == 0) {
1773    Diag(FExpr->getLocStart(), diag::warn_empty_format_string)
1774    << OrigFormatExpr->getSourceRange();
1775    return;
1776  }
1777
1778  if (isPrintf) {
1779    CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1780                         TheCall->getNumArgs() - firstDataArg,
1781                         isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1782                         HasVAListArg, TheCall, format_idx);
1783
1784    if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen))
1785      H.DoneProcessing();
1786  }
1787  else {
1788    CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1789                        TheCall->getNumArgs() - firstDataArg,
1790                        isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1791                        HasVAListArg, TheCall, format_idx);
1792
1793    if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen))
1794      H.DoneProcessing();
1795  }
1796}
1797
1798//===--- CHECK: Return Address of Stack Variable --------------------------===//
1799
1800static DeclRefExpr* EvalVal(Expr *E);
1801static DeclRefExpr* EvalAddr(Expr* E);
1802
1803/// CheckReturnStackAddr - Check if a return statement returns the address
1804///   of a stack variable.
1805void
1806Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
1807                           SourceLocation ReturnLoc) {
1808
1809  // Perform checking for returned stack addresses.
1810  if (lhsType->isPointerType() || lhsType->isBlockPointerType()) {
1811    if (DeclRefExpr *DR = EvalAddr(RetValExp))
1812      Diag(DR->getLocStart(), diag::warn_ret_stack_addr)
1813       << DR->getDecl()->getDeclName() << RetValExp->getSourceRange();
1814
1815    // Skip over implicit cast expressions when checking for block expressions.
1816    RetValExp = RetValExp->IgnoreParenCasts();
1817
1818    if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp))
1819      if (C->hasBlockDeclRefExprs())
1820        Diag(C->getLocStart(), diag::err_ret_local_block)
1821          << C->getSourceRange();
1822
1823    if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp))
1824      Diag(ALE->getLocStart(), diag::warn_ret_addr_label)
1825        << ALE->getSourceRange();
1826
1827  } else if (lhsType->isReferenceType()) {
1828    // Perform checking for stack values returned by reference.
1829    // Check for a reference to the stack
1830    if (DeclRefExpr *DR = EvalVal(RetValExp))
1831      Diag(DR->getLocStart(), diag::warn_ret_stack_ref)
1832        << DR->getDecl()->getDeclName() << RetValExp->getSourceRange();
1833  }
1834}
1835
1836/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
1837///  check if the expression in a return statement evaluates to an address
1838///  to a location on the stack.  The recursion is used to traverse the
1839///  AST of the return expression, with recursion backtracking when we
1840///  encounter a subexpression that (1) clearly does not lead to the address
1841///  of a stack variable or (2) is something we cannot determine leads to
1842///  the address of a stack variable based on such local checking.
1843///
1844///  EvalAddr processes expressions that are pointers that are used as
1845///  references (and not L-values).  EvalVal handles all other values.
1846///  At the base case of the recursion is a check for a DeclRefExpr* in
1847///  the refers to a stack variable.
1848///
1849///  This implementation handles:
1850///
1851///   * pointer-to-pointer casts
1852///   * implicit conversions from array references to pointers
1853///   * taking the address of fields
1854///   * arbitrary interplay between "&" and "*" operators
1855///   * pointer arithmetic from an address of a stack variable
1856///   * taking the address of an array element where the array is on the stack
1857static DeclRefExpr* EvalAddr(Expr *E) {
1858  // We should only be called for evaluating pointer expressions.
1859  assert((E->getType()->isAnyPointerType() ||
1860          E->getType()->isBlockPointerType() ||
1861          E->getType()->isObjCQualifiedIdType()) &&
1862         "EvalAddr only works on pointers");
1863
1864  // Our "symbolic interpreter" is just a dispatch off the currently
1865  // viewed AST node.  We then recursively traverse the AST by calling
1866  // EvalAddr and EvalVal appropriately.
1867  switch (E->getStmtClass()) {
1868  case Stmt::ParenExprClass:
1869    // Ignore parentheses.
1870    return EvalAddr(cast<ParenExpr>(E)->getSubExpr());
1871
1872  case Stmt::UnaryOperatorClass: {
1873    // The only unary operator that make sense to handle here
1874    // is AddrOf.  All others don't make sense as pointers.
1875    UnaryOperator *U = cast<UnaryOperator>(E);
1876
1877    if (U->getOpcode() == UnaryOperator::AddrOf)
1878      return EvalVal(U->getSubExpr());
1879    else
1880      return NULL;
1881  }
1882
1883  case Stmt::BinaryOperatorClass: {
1884    // Handle pointer arithmetic.  All other binary operators are not valid
1885    // in this context.
1886    BinaryOperator *B = cast<BinaryOperator>(E);
1887    BinaryOperator::Opcode op = B->getOpcode();
1888
1889    if (op != BinaryOperator::Add && op != BinaryOperator::Sub)
1890      return NULL;
1891
1892    Expr *Base = B->getLHS();
1893
1894    // Determine which argument is the real pointer base.  It could be
1895    // the RHS argument instead of the LHS.
1896    if (!Base->getType()->isPointerType()) Base = B->getRHS();
1897
1898    assert (Base->getType()->isPointerType());
1899    return EvalAddr(Base);
1900  }
1901
1902  // For conditional operators we need to see if either the LHS or RHS are
1903  // valid DeclRefExpr*s.  If one of them is valid, we return it.
1904  case Stmt::ConditionalOperatorClass: {
1905    ConditionalOperator *C = cast<ConditionalOperator>(E);
1906
1907    // Handle the GNU extension for missing LHS.
1908    if (Expr *lhsExpr = C->getLHS())
1909      if (DeclRefExpr* LHS = EvalAddr(lhsExpr))
1910        return LHS;
1911
1912     return EvalAddr(C->getRHS());
1913  }
1914
1915  // For casts, we need to handle conversions from arrays to
1916  // pointer values, and pointer-to-pointer conversions.
1917  case Stmt::ImplicitCastExprClass:
1918  case Stmt::CStyleCastExprClass:
1919  case Stmt::CXXFunctionalCastExprClass: {
1920    Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
1921    QualType T = SubExpr->getType();
1922
1923    if (SubExpr->getType()->isPointerType() ||
1924        SubExpr->getType()->isBlockPointerType() ||
1925        SubExpr->getType()->isObjCQualifiedIdType())
1926      return EvalAddr(SubExpr);
1927    else if (T->isArrayType())
1928      return EvalVal(SubExpr);
1929    else
1930      return 0;
1931  }
1932
1933  // C++ casts.  For dynamic casts, static casts, and const casts, we
1934  // are always converting from a pointer-to-pointer, so we just blow
1935  // through the cast.  In the case the dynamic cast doesn't fail (and
1936  // return NULL), we take the conservative route and report cases
1937  // where we return the address of a stack variable.  For Reinterpre
1938  // FIXME: The comment about is wrong; we're not always converting
1939  // from pointer to pointer. I'm guessing that this code should also
1940  // handle references to objects.
1941  case Stmt::CXXStaticCastExprClass:
1942  case Stmt::CXXDynamicCastExprClass:
1943  case Stmt::CXXConstCastExprClass:
1944  case Stmt::CXXReinterpretCastExprClass: {
1945      Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr();
1946      if (S->getType()->isPointerType() || S->getType()->isBlockPointerType())
1947        return EvalAddr(S);
1948      else
1949        return NULL;
1950  }
1951
1952  // Everything else: we simply don't reason about them.
1953  default:
1954    return NULL;
1955  }
1956}
1957
1958
1959///  EvalVal - This function is complements EvalAddr in the mutual recursion.
1960///   See the comments for EvalAddr for more details.
1961static DeclRefExpr* EvalVal(Expr *E) {
1962
1963  // We should only be called for evaluating non-pointer expressions, or
1964  // expressions with a pointer type that are not used as references but instead
1965  // are l-values (e.g., DeclRefExpr with a pointer type).
1966
1967  // Our "symbolic interpreter" is just a dispatch off the currently
1968  // viewed AST node.  We then recursively traverse the AST by calling
1969  // EvalAddr and EvalVal appropriately.
1970  switch (E->getStmtClass()) {
1971  case Stmt::DeclRefExprClass: {
1972    // DeclRefExpr: the base case.  When we hit a DeclRefExpr we are looking
1973    //  at code that refers to a variable's name.  We check if it has local
1974    //  storage within the function, and if so, return the expression.
1975    DeclRefExpr *DR = cast<DeclRefExpr>(E);
1976
1977    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
1978      if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR;
1979
1980    return NULL;
1981  }
1982
1983  case Stmt::ParenExprClass:
1984    // Ignore parentheses.
1985    return EvalVal(cast<ParenExpr>(E)->getSubExpr());
1986
1987  case Stmt::UnaryOperatorClass: {
1988    // The only unary operator that make sense to handle here
1989    // is Deref.  All others don't resolve to a "name."  This includes
1990    // handling all sorts of rvalues passed to a unary operator.
1991    UnaryOperator *U = cast<UnaryOperator>(E);
1992
1993    if (U->getOpcode() == UnaryOperator::Deref)
1994      return EvalAddr(U->getSubExpr());
1995
1996    return NULL;
1997  }
1998
1999  case Stmt::ArraySubscriptExprClass: {
2000    // Array subscripts are potential references to data on the stack.  We
2001    // retrieve the DeclRefExpr* for the array variable if it indeed
2002    // has local storage.
2003    return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase());
2004  }
2005
2006  case Stmt::ConditionalOperatorClass: {
2007    // For conditional operators we need to see if either the LHS or RHS are
2008    // non-NULL DeclRefExpr's.  If one is non-NULL, we return it.
2009    ConditionalOperator *C = cast<ConditionalOperator>(E);
2010
2011    // Handle the GNU extension for missing LHS.
2012    if (Expr *lhsExpr = C->getLHS())
2013      if (DeclRefExpr *LHS = EvalVal(lhsExpr))
2014        return LHS;
2015
2016    return EvalVal(C->getRHS());
2017  }
2018
2019  // Accesses to members are potential references to data on the stack.
2020  case Stmt::MemberExprClass: {
2021    MemberExpr *M = cast<MemberExpr>(E);
2022
2023    // Check for indirect access.  We only want direct field accesses.
2024    if (!M->isArrow())
2025      return EvalVal(M->getBase());
2026    else
2027      return NULL;
2028  }
2029
2030  // Everything else: we simply don't reason about them.
2031  default:
2032    return NULL;
2033  }
2034}
2035
2036//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
2037
2038/// Check for comparisons of floating point operands using != and ==.
2039/// Issue a warning if these are no self-comparisons, as they are not likely
2040/// to do what the programmer intended.
2041void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) {
2042  bool EmitWarning = true;
2043
2044  Expr* LeftExprSansParen = lex->IgnoreParens();
2045  Expr* RightExprSansParen = rex->IgnoreParens();
2046
2047  // Special case: check for x == x (which is OK).
2048  // Do not emit warnings for such cases.
2049  if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
2050    if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
2051      if (DRL->getDecl() == DRR->getDecl())
2052        EmitWarning = false;
2053
2054
2055  // Special case: check for comparisons against literals that can be exactly
2056  //  represented by APFloat.  In such cases, do not emit a warning.  This
2057  //  is a heuristic: often comparison against such literals are used to
2058  //  detect if a value in a variable has not changed.  This clearly can
2059  //  lead to false negatives.
2060  if (EmitWarning) {
2061    if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
2062      if (FLL->isExact())
2063        EmitWarning = false;
2064    } else
2065      if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){
2066        if (FLR->isExact())
2067          EmitWarning = false;
2068    }
2069  }
2070
2071  // Check for comparisons with builtin types.
2072  if (EmitWarning)
2073    if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
2074      if (CL->isBuiltinCall(Context))
2075        EmitWarning = false;
2076
2077  if (EmitWarning)
2078    if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
2079      if (CR->isBuiltinCall(Context))
2080        EmitWarning = false;
2081
2082  // Emit the diagnostic.
2083  if (EmitWarning)
2084    Diag(loc, diag::warn_floatingpoint_eq)
2085      << lex->getSourceRange() << rex->getSourceRange();
2086}
2087
2088//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===//
2089//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===//
2090
2091namespace {
2092
2093/// Structure recording the 'active' range of an integer-valued
2094/// expression.
2095struct IntRange {
2096  /// The number of bits active in the int.
2097  unsigned Width;
2098
2099  /// True if the int is known not to have negative values.
2100  bool NonNegative;
2101
2102  IntRange() {}
2103  IntRange(unsigned Width, bool NonNegative)
2104    : Width(Width), NonNegative(NonNegative)
2105  {}
2106
2107  // Returns the range of the bool type.
2108  static IntRange forBoolType() {
2109    return IntRange(1, true);
2110  }
2111
2112  // Returns the range of an integral type.
2113  static IntRange forType(ASTContext &C, QualType T) {
2114    return forCanonicalType(C, T->getCanonicalTypeInternal().getTypePtr());
2115  }
2116
2117  // Returns the range of an integeral type based on its canonical
2118  // representation.
2119  static IntRange forCanonicalType(ASTContext &C, const Type *T) {
2120    assert(T->isCanonicalUnqualified());
2121
2122    if (const VectorType *VT = dyn_cast<VectorType>(T))
2123      T = VT->getElementType().getTypePtr();
2124    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2125      T = CT->getElementType().getTypePtr();
2126
2127    if (const EnumType *ET = dyn_cast<EnumType>(T)) {
2128      EnumDecl *Enum = ET->getDecl();
2129      unsigned NumPositive = Enum->getNumPositiveBits();
2130      unsigned NumNegative = Enum->getNumNegativeBits();
2131
2132      return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0);
2133    }
2134
2135    const BuiltinType *BT = cast<BuiltinType>(T);
2136    assert(BT->isInteger());
2137
2138    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2139  }
2140
2141  // Returns the supremum of two ranges: i.e. their conservative merge.
2142  static IntRange join(IntRange L, IntRange R) {
2143    return IntRange(std::max(L.Width, R.Width),
2144                    L.NonNegative && R.NonNegative);
2145  }
2146
2147  // Returns the infinum of two ranges: i.e. their aggressive merge.
2148  static IntRange meet(IntRange L, IntRange R) {
2149    return IntRange(std::min(L.Width, R.Width),
2150                    L.NonNegative || R.NonNegative);
2151  }
2152};
2153
2154IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) {
2155  if (value.isSigned() && value.isNegative())
2156    return IntRange(value.getMinSignedBits(), false);
2157
2158  if (value.getBitWidth() > MaxWidth)
2159    value.trunc(MaxWidth);
2160
2161  // isNonNegative() just checks the sign bit without considering
2162  // signedness.
2163  return IntRange(value.getActiveBits(), true);
2164}
2165
2166IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty,
2167                       unsigned MaxWidth) {
2168  if (result.isInt())
2169    return GetValueRange(C, result.getInt(), MaxWidth);
2170
2171  if (result.isVector()) {
2172    IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth);
2173    for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) {
2174      IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth);
2175      R = IntRange::join(R, El);
2176    }
2177    return R;
2178  }
2179
2180  if (result.isComplexInt()) {
2181    IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth);
2182    IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth);
2183    return IntRange::join(R, I);
2184  }
2185
2186  // This can happen with lossless casts to intptr_t of "based" lvalues.
2187  // Assume it might use arbitrary bits.
2188  // FIXME: The only reason we need to pass the type in here is to get
2189  // the sign right on this one case.  It would be nice if APValue
2190  // preserved this.
2191  assert(result.isLValue());
2192  return IntRange(MaxWidth, Ty->isUnsignedIntegerType());
2193}
2194
2195/// Pseudo-evaluate the given integer expression, estimating the
2196/// range of values it might take.
2197///
2198/// \param MaxWidth - the width to which the value will be truncated
2199IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) {
2200  E = E->IgnoreParens();
2201
2202  // Try a full evaluation first.
2203  Expr::EvalResult result;
2204  if (E->Evaluate(result, C))
2205    return GetValueRange(C, result.Val, E->getType(), MaxWidth);
2206
2207  // I think we only want to look through implicit casts here; if the
2208  // user has an explicit widening cast, we should treat the value as
2209  // being of the new, wider type.
2210  if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) {
2211    if (CE->getCastKind() == CastExpr::CK_NoOp)
2212      return GetExprRange(C, CE->getSubExpr(), MaxWidth);
2213
2214    IntRange OutputTypeRange = IntRange::forType(C, CE->getType());
2215
2216    bool isIntegerCast = (CE->getCastKind() == CastExpr::CK_IntegralCast);
2217    if (!isIntegerCast && CE->getCastKind() == CastExpr::CK_Unknown)
2218      isIntegerCast = CE->getSubExpr()->getType()->isIntegerType();
2219
2220    // Assume that non-integer casts can span the full range of the type.
2221    if (!isIntegerCast)
2222      return OutputTypeRange;
2223
2224    IntRange SubRange
2225      = GetExprRange(C, CE->getSubExpr(),
2226                     std::min(MaxWidth, OutputTypeRange.Width));
2227
2228    // Bail out if the subexpr's range is as wide as the cast type.
2229    if (SubRange.Width >= OutputTypeRange.Width)
2230      return OutputTypeRange;
2231
2232    // Otherwise, we take the smaller width, and we're non-negative if
2233    // either the output type or the subexpr is.
2234    return IntRange(SubRange.Width,
2235                    SubRange.NonNegative || OutputTypeRange.NonNegative);
2236  }
2237
2238  if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
2239    // If we can fold the condition, just take that operand.
2240    bool CondResult;
2241    if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C))
2242      return GetExprRange(C, CondResult ? CO->getTrueExpr()
2243                                        : CO->getFalseExpr(),
2244                          MaxWidth);
2245
2246    // Otherwise, conservatively merge.
2247    IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth);
2248    IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth);
2249    return IntRange::join(L, R);
2250  }
2251
2252  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
2253    switch (BO->getOpcode()) {
2254
2255    // Boolean-valued operations are single-bit and positive.
2256    case BinaryOperator::LAnd:
2257    case BinaryOperator::LOr:
2258    case BinaryOperator::LT:
2259    case BinaryOperator::GT:
2260    case BinaryOperator::LE:
2261    case BinaryOperator::GE:
2262    case BinaryOperator::EQ:
2263    case BinaryOperator::NE:
2264      return IntRange::forBoolType();
2265
2266    // The type of these compound assignments is the type of the LHS,
2267    // so the RHS is not necessarily an integer.
2268    case BinaryOperator::MulAssign:
2269    case BinaryOperator::DivAssign:
2270    case BinaryOperator::RemAssign:
2271    case BinaryOperator::AddAssign:
2272    case BinaryOperator::SubAssign:
2273      return IntRange::forType(C, E->getType());
2274
2275    // Operations with opaque sources are black-listed.
2276    case BinaryOperator::PtrMemD:
2277    case BinaryOperator::PtrMemI:
2278      return IntRange::forType(C, E->getType());
2279
2280    // Bitwise-and uses the *infinum* of the two source ranges.
2281    case BinaryOperator::And:
2282    case BinaryOperator::AndAssign:
2283      return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth),
2284                            GetExprRange(C, BO->getRHS(), MaxWidth));
2285
2286    // Left shift gets black-listed based on a judgement call.
2287    case BinaryOperator::Shl:
2288      // ...except that we want to treat '1 << (blah)' as logically
2289      // positive.  It's an important idiom.
2290      if (IntegerLiteral *I
2291            = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) {
2292        if (I->getValue() == 1) {
2293          IntRange R = IntRange::forType(C, E->getType());
2294          return IntRange(R.Width, /*NonNegative*/ true);
2295        }
2296      }
2297      // fallthrough
2298
2299    case BinaryOperator::ShlAssign:
2300      return IntRange::forType(C, E->getType());
2301
2302    // Right shift by a constant can narrow its left argument.
2303    case BinaryOperator::Shr:
2304    case BinaryOperator::ShrAssign: {
2305      IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2306
2307      // If the shift amount is a positive constant, drop the width by
2308      // that much.
2309      llvm::APSInt shift;
2310      if (BO->getRHS()->isIntegerConstantExpr(shift, C) &&
2311          shift.isNonNegative()) {
2312        unsigned zext = shift.getZExtValue();
2313        if (zext >= L.Width)
2314          L.Width = (L.NonNegative ? 0 : 1);
2315        else
2316          L.Width -= zext;
2317      }
2318
2319      return L;
2320    }
2321
2322    // Comma acts as its right operand.
2323    case BinaryOperator::Comma:
2324      return GetExprRange(C, BO->getRHS(), MaxWidth);
2325
2326    // Black-list pointer subtractions.
2327    case BinaryOperator::Sub:
2328      if (BO->getLHS()->getType()->isPointerType())
2329        return IntRange::forType(C, E->getType());
2330      // fallthrough
2331
2332    default:
2333      break;
2334    }
2335
2336    // Treat every other operator as if it were closed on the
2337    // narrowest type that encompasses both operands.
2338    IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2339    IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth);
2340    return IntRange::join(L, R);
2341  }
2342
2343  if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
2344    switch (UO->getOpcode()) {
2345    // Boolean-valued operations are white-listed.
2346    case UnaryOperator::LNot:
2347      return IntRange::forBoolType();
2348
2349    // Operations with opaque sources are black-listed.
2350    case UnaryOperator::Deref:
2351    case UnaryOperator::AddrOf: // should be impossible
2352    case UnaryOperator::OffsetOf:
2353      return IntRange::forType(C, E->getType());
2354
2355    default:
2356      return GetExprRange(C, UO->getSubExpr(), MaxWidth);
2357    }
2358  }
2359
2360  if (dyn_cast<OffsetOfExpr>(E)) {
2361    IntRange::forType(C, E->getType());
2362  }
2363
2364  FieldDecl *BitField = E->getBitField();
2365  if (BitField) {
2366    llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C);
2367    unsigned BitWidth = BitWidthAP.getZExtValue();
2368
2369    return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType());
2370  }
2371
2372  return IntRange::forType(C, E->getType());
2373}
2374
2375IntRange GetExprRange(ASTContext &C, Expr *E) {
2376  return GetExprRange(C, E, C.getIntWidth(E->getType()));
2377}
2378
2379/// Checks whether the given value, which currently has the given
2380/// source semantics, has the same value when coerced through the
2381/// target semantics.
2382bool IsSameFloatAfterCast(const llvm::APFloat &value,
2383                          const llvm::fltSemantics &Src,
2384                          const llvm::fltSemantics &Tgt) {
2385  llvm::APFloat truncated = value;
2386
2387  bool ignored;
2388  truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored);
2389  truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored);
2390
2391  return truncated.bitwiseIsEqual(value);
2392}
2393
2394/// Checks whether the given value, which currently has the given
2395/// source semantics, has the same value when coerced through the
2396/// target semantics.
2397///
2398/// The value might be a vector of floats (or a complex number).
2399bool IsSameFloatAfterCast(const APValue &value,
2400                          const llvm::fltSemantics &Src,
2401                          const llvm::fltSemantics &Tgt) {
2402  if (value.isFloat())
2403    return IsSameFloatAfterCast(value.getFloat(), Src, Tgt);
2404
2405  if (value.isVector()) {
2406    for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i)
2407      if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt))
2408        return false;
2409    return true;
2410  }
2411
2412  assert(value.isComplexFloat());
2413  return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) &&
2414          IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt));
2415}
2416
2417void AnalyzeImplicitConversions(Sema &S, Expr *E);
2418
2419bool IsZero(Sema &S, Expr *E) {
2420  llvm::APSInt Value;
2421  return E->isIntegerConstantExpr(Value, S.Context) && Value == 0;
2422}
2423
2424void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) {
2425  BinaryOperator::Opcode op = E->getOpcode();
2426  if (op == BinaryOperator::LT && IsZero(S, E->getRHS())) {
2427    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2428      << "< 0" << "false"
2429      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2430  } else if (op == BinaryOperator::GE && IsZero(S, E->getRHS())) {
2431    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2432      << ">= 0" << "true"
2433      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2434  } else if (op == BinaryOperator::GT && IsZero(S, E->getLHS())) {
2435    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
2436      << "0 >" << "false"
2437      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2438  } else if (op == BinaryOperator::LE && IsZero(S, E->getLHS())) {
2439    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
2440      << "0 <=" << "true"
2441      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2442  }
2443}
2444
2445/// Analyze the operands of the given comparison.  Implements the
2446/// fallback case from AnalyzeComparison.
2447void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) {
2448  AnalyzeImplicitConversions(S, E->getLHS());
2449  AnalyzeImplicitConversions(S, E->getRHS());
2450}
2451
2452/// \brief Implements -Wsign-compare.
2453///
2454/// \param lex the left-hand expression
2455/// \param rex the right-hand expression
2456/// \param OpLoc the location of the joining operator
2457/// \param BinOpc binary opcode or 0
2458void AnalyzeComparison(Sema &S, BinaryOperator *E) {
2459  // The type the comparison is being performed in.
2460  QualType T = E->getLHS()->getType();
2461  assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType())
2462         && "comparison with mismatched types");
2463
2464  // We don't do anything special if this isn't an unsigned integral
2465  // comparison:  we're only interested in integral comparisons, and
2466  // signed comparisons only happen in cases we don't care to warn about.
2467  if (!T->hasUnsignedIntegerRepresentation())
2468    return AnalyzeImpConvsInComparison(S, E);
2469
2470  Expr *lex = E->getLHS()->IgnoreParenImpCasts();
2471  Expr *rex = E->getRHS()->IgnoreParenImpCasts();
2472
2473  // Check to see if one of the (unmodified) operands is of different
2474  // signedness.
2475  Expr *signedOperand, *unsignedOperand;
2476  if (lex->getType()->hasSignedIntegerRepresentation()) {
2477    assert(!rex->getType()->hasSignedIntegerRepresentation() &&
2478           "unsigned comparison between two signed integer expressions?");
2479    signedOperand = lex;
2480    unsignedOperand = rex;
2481  } else if (rex->getType()->hasSignedIntegerRepresentation()) {
2482    signedOperand = rex;
2483    unsignedOperand = lex;
2484  } else {
2485    CheckTrivialUnsignedComparison(S, E);
2486    return AnalyzeImpConvsInComparison(S, E);
2487  }
2488
2489  // Otherwise, calculate the effective range of the signed operand.
2490  IntRange signedRange = GetExprRange(S.Context, signedOperand);
2491
2492  // Go ahead and analyze implicit conversions in the operands.  Note
2493  // that we skip the implicit conversions on both sides.
2494  AnalyzeImplicitConversions(S, lex);
2495  AnalyzeImplicitConversions(S, rex);
2496
2497  // If the signed range is non-negative, -Wsign-compare won't fire,
2498  // but we should still check for comparisons which are always true
2499  // or false.
2500  if (signedRange.NonNegative)
2501    return CheckTrivialUnsignedComparison(S, E);
2502
2503  // For (in)equality comparisons, if the unsigned operand is a
2504  // constant which cannot collide with a overflowed signed operand,
2505  // then reinterpreting the signed operand as unsigned will not
2506  // change the result of the comparison.
2507  if (E->isEqualityOp()) {
2508    unsigned comparisonWidth = S.Context.getIntWidth(T);
2509    IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand);
2510
2511    // We should never be unable to prove that the unsigned operand is
2512    // non-negative.
2513    assert(unsignedRange.NonNegative && "unsigned range includes negative?");
2514
2515    if (unsignedRange.Width < comparisonWidth)
2516      return;
2517  }
2518
2519  S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison)
2520    << lex->getType() << rex->getType()
2521    << lex->getSourceRange() << rex->getSourceRange();
2522}
2523
2524/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
2525void DiagnoseImpCast(Sema &S, Expr *E, QualType T, unsigned diag) {
2526  S.Diag(E->getExprLoc(), diag) << E->getType() << T << E->getSourceRange();
2527}
2528
2529void CheckImplicitConversion(Sema &S, Expr *E, QualType T,
2530                             bool *ICContext = 0) {
2531  if (E->isTypeDependent() || E->isValueDependent()) return;
2532
2533  const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr();
2534  const Type *Target = S.Context.getCanonicalType(T).getTypePtr();
2535  if (Source == Target) return;
2536  if (Target->isDependentType()) return;
2537
2538  // Never diagnose implicit casts to bool.
2539  if (Target->isSpecificBuiltinType(BuiltinType::Bool))
2540    return;
2541
2542  // Strip vector types.
2543  if (isa<VectorType>(Source)) {
2544    if (!isa<VectorType>(Target))
2545      return DiagnoseImpCast(S, E, T, diag::warn_impcast_vector_scalar);
2546
2547    Source = cast<VectorType>(Source)->getElementType().getTypePtr();
2548    Target = cast<VectorType>(Target)->getElementType().getTypePtr();
2549  }
2550
2551  // Strip complex types.
2552  if (isa<ComplexType>(Source)) {
2553    if (!isa<ComplexType>(Target))
2554      return DiagnoseImpCast(S, E, T, diag::warn_impcast_complex_scalar);
2555
2556    Source = cast<ComplexType>(Source)->getElementType().getTypePtr();
2557    Target = cast<ComplexType>(Target)->getElementType().getTypePtr();
2558  }
2559
2560  const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source);
2561  const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target);
2562
2563  // If the source is floating point...
2564  if (SourceBT && SourceBT->isFloatingPoint()) {
2565    // ...and the target is floating point...
2566    if (TargetBT && TargetBT->isFloatingPoint()) {
2567      // ...then warn if we're dropping FP rank.
2568
2569      // Builtin FP kinds are ordered by increasing FP rank.
2570      if (SourceBT->getKind() > TargetBT->getKind()) {
2571        // Don't warn about float constants that are precisely
2572        // representable in the target type.
2573        Expr::EvalResult result;
2574        if (E->Evaluate(result, S.Context)) {
2575          // Value might be a float, a float vector, or a float complex.
2576          if (IsSameFloatAfterCast(result.Val,
2577                   S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)),
2578                   S.Context.getFloatTypeSemantics(QualType(SourceBT, 0))))
2579            return;
2580        }
2581
2582        DiagnoseImpCast(S, E, T, diag::warn_impcast_float_precision);
2583      }
2584      return;
2585    }
2586
2587    // If the target is integral, always warn.
2588    if ((TargetBT && TargetBT->isInteger()))
2589      // TODO: don't warn for integer values?
2590      DiagnoseImpCast(S, E, T, diag::warn_impcast_float_integer);
2591
2592    return;
2593  }
2594
2595  if (!Source->isIntegerType() || !Target->isIntegerType())
2596    return;
2597
2598  IntRange SourceRange = GetExprRange(S.Context, E);
2599  IntRange TargetRange = IntRange::forCanonicalType(S.Context, Target);
2600
2601  if (SourceRange.Width > TargetRange.Width) {
2602    // People want to build with -Wshorten-64-to-32 and not -Wconversion
2603    // and by god we'll let them.
2604    if (SourceRange.Width == 64 && TargetRange.Width == 32)
2605      return DiagnoseImpCast(S, E, T, diag::warn_impcast_integer_64_32);
2606    return DiagnoseImpCast(S, E, T, diag::warn_impcast_integer_precision);
2607  }
2608
2609  if ((TargetRange.NonNegative && !SourceRange.NonNegative) ||
2610      (!TargetRange.NonNegative && SourceRange.NonNegative &&
2611       SourceRange.Width == TargetRange.Width)) {
2612    unsigned DiagID = diag::warn_impcast_integer_sign;
2613
2614    // Traditionally, gcc has warned about this under -Wsign-compare.
2615    // We also want to warn about it in -Wconversion.
2616    // So if -Wconversion is off, use a completely identical diagnostic
2617    // in the sign-compare group.
2618    // The conditional-checking code will
2619    if (ICContext) {
2620      DiagID = diag::warn_impcast_integer_sign_conditional;
2621      *ICContext = true;
2622    }
2623
2624    return DiagnoseImpCast(S, E, T, DiagID);
2625  }
2626
2627  return;
2628}
2629
2630void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T);
2631
2632void CheckConditionalOperand(Sema &S, Expr *E, QualType T,
2633                             bool &ICContext) {
2634  E = E->IgnoreParenImpCasts();
2635
2636  if (isa<ConditionalOperator>(E))
2637    return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T);
2638
2639  AnalyzeImplicitConversions(S, E);
2640  if (E->getType() != T)
2641    return CheckImplicitConversion(S, E, T, &ICContext);
2642  return;
2643}
2644
2645void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) {
2646  AnalyzeImplicitConversions(S, E->getCond());
2647
2648  bool Suspicious = false;
2649  CheckConditionalOperand(S, E->getTrueExpr(), T, Suspicious);
2650  CheckConditionalOperand(S, E->getFalseExpr(), T, Suspicious);
2651
2652  // If -Wconversion would have warned about either of the candidates
2653  // for a signedness conversion to the context type...
2654  if (!Suspicious) return;
2655
2656  // ...but it's currently ignored...
2657  if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional))
2658    return;
2659
2660  // ...and -Wsign-compare isn't...
2661  if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional))
2662    return;
2663
2664  // ...then check whether it would have warned about either of the
2665  // candidates for a signedness conversion to the condition type.
2666  if (E->getType() != T) {
2667    Suspicious = false;
2668    CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(),
2669                            E->getType(), &Suspicious);
2670    if (!Suspicious)
2671      CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(),
2672                              E->getType(), &Suspicious);
2673    if (!Suspicious)
2674      return;
2675  }
2676
2677  // If so, emit a diagnostic under -Wsign-compare.
2678  Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts();
2679  Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts();
2680  S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional)
2681    << lex->getType() << rex->getType()
2682    << lex->getSourceRange() << rex->getSourceRange();
2683}
2684
2685/// AnalyzeImplicitConversions - Find and report any interesting
2686/// implicit conversions in the given expression.  There are a couple
2687/// of competing diagnostics here, -Wconversion and -Wsign-compare.
2688void AnalyzeImplicitConversions(Sema &S, Expr *OrigE) {
2689  QualType T = OrigE->getType();
2690  Expr *E = OrigE->IgnoreParenImpCasts();
2691
2692  // For conditional operators, we analyze the arguments as if they
2693  // were being fed directly into the output.
2694  if (isa<ConditionalOperator>(E)) {
2695    ConditionalOperator *CO = cast<ConditionalOperator>(E);
2696    CheckConditionalOperator(S, CO, T);
2697    return;
2698  }
2699
2700  // Go ahead and check any implicit conversions we might have skipped.
2701  // The non-canonical typecheck is just an optimization;
2702  // CheckImplicitConversion will filter out dead implicit conversions.
2703  if (E->getType() != T)
2704    CheckImplicitConversion(S, E, T);
2705
2706  // Now continue drilling into this expression.
2707
2708  // Skip past explicit casts.
2709  if (isa<ExplicitCastExpr>(E)) {
2710    E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts();
2711    return AnalyzeImplicitConversions(S, E);
2712  }
2713
2714  // Do a somewhat different check with comparison operators.
2715  if (isa<BinaryOperator>(E) && cast<BinaryOperator>(E)->isComparisonOp())
2716    return AnalyzeComparison(S, cast<BinaryOperator>(E));
2717
2718  // These break the otherwise-useful invariant below.  Fortunately,
2719  // we don't really need to recurse into them, because any internal
2720  // expressions should have been analyzed already when they were
2721  // built into statements.
2722  if (isa<StmtExpr>(E)) return;
2723
2724  // Don't descend into unevaluated contexts.
2725  if (isa<SizeOfAlignOfExpr>(E)) return;
2726
2727  // Now just recurse over the expression's children.
2728  for (Stmt::child_iterator I = E->child_begin(), IE = E->child_end();
2729         I != IE; ++I)
2730    AnalyzeImplicitConversions(S, cast<Expr>(*I));
2731}
2732
2733} // end anonymous namespace
2734
2735/// Diagnoses "dangerous" implicit conversions within the given
2736/// expression (which is a full expression).  Implements -Wconversion
2737/// and -Wsign-compare.
2738void Sema::CheckImplicitConversions(Expr *E) {
2739  // Don't diagnose in unevaluated contexts.
2740  if (ExprEvalContexts.back().Context == Sema::Unevaluated)
2741    return;
2742
2743  // Don't diagnose for value- or type-dependent expressions.
2744  if (E->isTypeDependent() || E->isValueDependent())
2745    return;
2746
2747  AnalyzeImplicitConversions(*this, E);
2748}
2749
2750/// CheckParmsForFunctionDef - Check that the parameters of the given
2751/// function are appropriate for the definition of a function. This
2752/// takes care of any checks that cannot be performed on the
2753/// declaration itself, e.g., that the types of each of the function
2754/// parameters are complete.
2755bool Sema::CheckParmsForFunctionDef(FunctionDecl *FD) {
2756  bool HasInvalidParm = false;
2757  for (unsigned p = 0, NumParams = FD->getNumParams(); p < NumParams; ++p) {
2758    ParmVarDecl *Param = FD->getParamDecl(p);
2759
2760    // C99 6.7.5.3p4: the parameters in a parameter type list in a
2761    // function declarator that is part of a function definition of
2762    // that function shall not have incomplete type.
2763    //
2764    // This is also C++ [dcl.fct]p6.
2765    if (!Param->isInvalidDecl() &&
2766        RequireCompleteType(Param->getLocation(), Param->getType(),
2767                               diag::err_typecheck_decl_incomplete_type)) {
2768      Param->setInvalidDecl();
2769      HasInvalidParm = true;
2770    }
2771
2772    // C99 6.9.1p5: If the declarator includes a parameter type list, the
2773    // declaration of each parameter shall include an identifier.
2774    if (Param->getIdentifier() == 0 &&
2775        !Param->isImplicit() &&
2776        !getLangOptions().CPlusPlus)
2777      Diag(Param->getLocation(), diag::err_parameter_name_omitted);
2778
2779    // C99 6.7.5.3p12:
2780    //   If the function declarator is not part of a definition of that
2781    //   function, parameters may have incomplete type and may use the [*]
2782    //   notation in their sequences of declarator specifiers to specify
2783    //   variable length array types.
2784    QualType PType = Param->getOriginalType();
2785    if (const ArrayType *AT = Context.getAsArrayType(PType)) {
2786      if (AT->getSizeModifier() == ArrayType::Star) {
2787        // FIXME: This diagnosic should point the the '[*]' if source-location
2788        // information is added for it.
2789        Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
2790      }
2791    }
2792  }
2793
2794  return HasInvalidParm;
2795}
2796