SemaChecking.cpp revision ff331c15729f7d4439d253c97f4d60f2a7ffd0c6
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}
1174
1175SourceRange CheckFormatHandler::getFormatStringRange() {
1176  return OrigFormatExpr->getSourceRange();
1177}
1178
1179CharSourceRange CheckFormatHandler::
1180getSpecifierRange(const char *startSpecifier, unsigned specifierLen) {
1181  SourceLocation Start = getLocationOfByte(startSpecifier);
1182  SourceLocation End   = getLocationOfByte(startSpecifier + specifierLen - 1);
1183
1184  // Advance the end SourceLocation by one due to half-open ranges.
1185  End = End.getFileLocWithOffset(1);
1186
1187  return CharSourceRange::getCharRange(Start, End);
1188}
1189
1190SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) {
1191  return S.getLocationOfStringLiteralByte(FExpr, x - Beg);
1192}
1193
1194void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier,
1195                                                   unsigned specifierLen){
1196  SourceLocation Loc = getLocationOfByte(startSpecifier);
1197  S.Diag(Loc, diag::warn_printf_incomplete_specifier)
1198    << getSpecifierRange(startSpecifier, specifierLen);
1199}
1200
1201void
1202CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen,
1203                                     analyze_format_string::PositionContext p) {
1204  SourceLocation Loc = getLocationOfByte(startPos);
1205  S.Diag(Loc, diag::warn_format_invalid_positional_specifier)
1206    << (unsigned) p << getSpecifierRange(startPos, posLen);
1207}
1208
1209void CheckFormatHandler::HandleZeroPosition(const char *startPos,
1210                                            unsigned posLen) {
1211  SourceLocation Loc = getLocationOfByte(startPos);
1212  S.Diag(Loc, diag::warn_format_zero_positional_specifier)
1213    << getSpecifierRange(startPos, posLen);
1214}
1215
1216void CheckFormatHandler::HandleNullChar(const char *nullCharacter) {
1217  // The presence of a null character is likely an error.
1218  S.Diag(getLocationOfByte(nullCharacter),
1219         diag::warn_printf_format_string_contains_null_char)
1220    << getFormatStringRange();
1221}
1222
1223const Expr *CheckFormatHandler::getDataArg(unsigned i) const {
1224  return TheCall->getArg(FirstDataArg + i);
1225}
1226
1227void CheckFormatHandler::DoneProcessing() {
1228    // Does the number of data arguments exceed the number of
1229    // format conversions in the format string?
1230  if (!HasVAListArg) {
1231      // Find any arguments that weren't covered.
1232    CoveredArgs.flip();
1233    signed notCoveredArg = CoveredArgs.find_first();
1234    if (notCoveredArg >= 0) {
1235      assert((unsigned)notCoveredArg < NumDataArgs);
1236      S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(),
1237             diag::warn_printf_data_arg_not_used)
1238      << getFormatStringRange();
1239    }
1240  }
1241}
1242
1243bool
1244CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex,
1245                                                     SourceLocation Loc,
1246                                                     const char *startSpec,
1247                                                     unsigned specifierLen,
1248                                                     const char *csStart,
1249                                                     unsigned csLen) {
1250
1251  bool keepGoing = true;
1252  if (argIndex < NumDataArgs) {
1253    // Consider the argument coverered, even though the specifier doesn't
1254    // make sense.
1255    CoveredArgs.set(argIndex);
1256  }
1257  else {
1258    // If argIndex exceeds the number of data arguments we
1259    // don't issue a warning because that is just a cascade of warnings (and
1260    // they may have intended '%%' anyway). We don't want to continue processing
1261    // the format string after this point, however, as we will like just get
1262    // gibberish when trying to match arguments.
1263    keepGoing = false;
1264  }
1265
1266  S.Diag(Loc, diag::warn_format_invalid_conversion)
1267    << llvm::StringRef(csStart, csLen)
1268    << getSpecifierRange(startSpec, specifierLen);
1269
1270  return keepGoing;
1271}
1272
1273//===--- CHECK: Printf format string checking ------------------------------===//
1274
1275namespace {
1276class CheckPrintfHandler : public CheckFormatHandler {
1277public:
1278  CheckPrintfHandler(Sema &s, const StringLiteral *fexpr,
1279                     const Expr *origFormatExpr, unsigned firstDataArg,
1280                     unsigned numDataArgs, bool isObjCLiteral,
1281                     const char *beg, bool hasVAListArg,
1282                     const CallExpr *theCall, unsigned formatIdx)
1283  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1284                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1285                       theCall, formatIdx) {}
1286
1287
1288  bool HandleInvalidPrintfConversionSpecifier(
1289                                      const analyze_printf::PrintfSpecifier &FS,
1290                                      const char *startSpecifier,
1291                                      unsigned specifierLen);
1292
1293  bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS,
1294                             const char *startSpecifier,
1295                             unsigned specifierLen);
1296
1297  bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k,
1298                    const char *startSpecifier, unsigned specifierLen);
1299  void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS,
1300                           const analyze_printf::OptionalAmount &Amt,
1301                           unsigned type,
1302                           const char *startSpecifier, unsigned specifierLen);
1303  void HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1304                  const analyze_printf::OptionalFlag &flag,
1305                  const char *startSpecifier, unsigned specifierLen);
1306  void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS,
1307                         const analyze_printf::OptionalFlag &ignoredFlag,
1308                         const analyze_printf::OptionalFlag &flag,
1309                         const char *startSpecifier, unsigned specifierLen);
1310};
1311}
1312
1313bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier(
1314                                      const analyze_printf::PrintfSpecifier &FS,
1315                                      const char *startSpecifier,
1316                                      unsigned specifierLen) {
1317  const analyze_printf::PrintfConversionSpecifier &CS =
1318    FS.getConversionSpecifier();
1319
1320  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1321                                          getLocationOfByte(CS.getStart()),
1322                                          startSpecifier, specifierLen,
1323                                          CS.getStart(), CS.getLength());
1324}
1325
1326bool CheckPrintfHandler::HandleAmount(
1327                               const analyze_format_string::OptionalAmount &Amt,
1328                               unsigned k, const char *startSpecifier,
1329                               unsigned specifierLen) {
1330
1331  if (Amt.hasDataArgument()) {
1332    if (!HasVAListArg) {
1333      unsigned argIndex = Amt.getArgIndex();
1334      if (argIndex >= NumDataArgs) {
1335        S.Diag(getLocationOfByte(Amt.getStart()),
1336               diag::warn_printf_asterisk_missing_arg)
1337          << k << getSpecifierRange(startSpecifier, specifierLen);
1338        // Don't do any more checking.  We will just emit
1339        // spurious errors.
1340        return false;
1341      }
1342
1343      // Type check the data argument.  It should be an 'int'.
1344      // Although not in conformance with C99, we also allow the argument to be
1345      // an 'unsigned int' as that is a reasonably safe case.  GCC also
1346      // doesn't emit a warning for that case.
1347      CoveredArgs.set(argIndex);
1348      const Expr *Arg = getDataArg(argIndex);
1349      QualType T = Arg->getType();
1350
1351      const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context);
1352      assert(ATR.isValid());
1353
1354      if (!ATR.matchesType(S.Context, T)) {
1355        S.Diag(getLocationOfByte(Amt.getStart()),
1356               diag::warn_printf_asterisk_wrong_type)
1357          << k
1358          << ATR.getRepresentativeType(S.Context) << T
1359          << getSpecifierRange(startSpecifier, specifierLen)
1360          << Arg->getSourceRange();
1361        // Don't do any more checking.  We will just emit
1362        // spurious errors.
1363        return false;
1364      }
1365    }
1366  }
1367  return true;
1368}
1369
1370void CheckPrintfHandler::HandleInvalidAmount(
1371                                      const analyze_printf::PrintfSpecifier &FS,
1372                                      const analyze_printf::OptionalAmount &Amt,
1373                                      unsigned type,
1374                                      const char *startSpecifier,
1375                                      unsigned specifierLen) {
1376  const analyze_printf::PrintfConversionSpecifier &CS =
1377    FS.getConversionSpecifier();
1378  switch (Amt.getHowSpecified()) {
1379  case analyze_printf::OptionalAmount::Constant:
1380    S.Diag(getLocationOfByte(Amt.getStart()),
1381        diag::warn_printf_nonsensical_optional_amount)
1382      << type
1383      << CS.toString()
1384      << getSpecifierRange(startSpecifier, specifierLen)
1385      << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(),
1386          Amt.getConstantLength()));
1387    break;
1388
1389  default:
1390    S.Diag(getLocationOfByte(Amt.getStart()),
1391        diag::warn_printf_nonsensical_optional_amount)
1392      << type
1393      << CS.toString()
1394      << getSpecifierRange(startSpecifier, specifierLen);
1395    break;
1396  }
1397}
1398
1399void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1400                                    const analyze_printf::OptionalFlag &flag,
1401                                    const char *startSpecifier,
1402                                    unsigned specifierLen) {
1403  // Warn about pointless flag with a fixit removal.
1404  const analyze_printf::PrintfConversionSpecifier &CS =
1405    FS.getConversionSpecifier();
1406  S.Diag(getLocationOfByte(flag.getPosition()),
1407      diag::warn_printf_nonsensical_flag)
1408    << flag.toString() << CS.toString()
1409    << getSpecifierRange(startSpecifier, specifierLen)
1410    << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1));
1411}
1412
1413void CheckPrintfHandler::HandleIgnoredFlag(
1414                                const analyze_printf::PrintfSpecifier &FS,
1415                                const analyze_printf::OptionalFlag &ignoredFlag,
1416                                const analyze_printf::OptionalFlag &flag,
1417                                const char *startSpecifier,
1418                                unsigned specifierLen) {
1419  // Warn about ignored flag with a fixit removal.
1420  S.Diag(getLocationOfByte(ignoredFlag.getPosition()),
1421      diag::warn_printf_ignored_flag)
1422    << ignoredFlag.toString() << flag.toString()
1423    << getSpecifierRange(startSpecifier, specifierLen)
1424    << FixItHint::CreateRemoval(getSpecifierRange(
1425        ignoredFlag.getPosition(), 1));
1426}
1427
1428bool
1429CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
1430                                            &FS,
1431                                          const char *startSpecifier,
1432                                          unsigned specifierLen) {
1433
1434  using namespace analyze_format_string;
1435  using namespace analyze_printf;
1436  const PrintfConversionSpecifier &CS = FS.getConversionSpecifier();
1437
1438  if (FS.consumesDataArgument()) {
1439    if (atFirstArg) {
1440        atFirstArg = false;
1441        usesPositionalArgs = FS.usesPositionalArg();
1442    }
1443    else if (usesPositionalArgs != FS.usesPositionalArg()) {
1444      // Cannot mix-and-match positional and non-positional arguments.
1445      S.Diag(getLocationOfByte(CS.getStart()),
1446             diag::warn_format_mix_positional_nonpositional_args)
1447        << getSpecifierRange(startSpecifier, specifierLen);
1448      return false;
1449    }
1450  }
1451
1452  // First check if the field width, precision, and conversion specifier
1453  // have matching data arguments.
1454  if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0,
1455                    startSpecifier, specifierLen)) {
1456    return false;
1457  }
1458
1459  if (!HandleAmount(FS.getPrecision(), /* precision */ 1,
1460                    startSpecifier, specifierLen)) {
1461    return false;
1462  }
1463
1464  if (!CS.consumesDataArgument()) {
1465    // FIXME: Technically specifying a precision or field width here
1466    // makes no sense.  Worth issuing a warning at some point.
1467    return true;
1468  }
1469
1470  // Consume the argument.
1471  unsigned argIndex = FS.getArgIndex();
1472  if (argIndex < NumDataArgs) {
1473    // The check to see if the argIndex is valid will come later.
1474    // We set the bit here because we may exit early from this
1475    // function if we encounter some other error.
1476    CoveredArgs.set(argIndex);
1477  }
1478
1479  // Check for using an Objective-C specific conversion specifier
1480  // in a non-ObjC literal.
1481  if (!IsObjCLiteral && CS.isObjCArg()) {
1482    return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier,
1483                                                  specifierLen);
1484  }
1485
1486  // Check for invalid use of field width
1487  if (!FS.hasValidFieldWidth()) {
1488    HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0,
1489        startSpecifier, specifierLen);
1490  }
1491
1492  // Check for invalid use of precision
1493  if (!FS.hasValidPrecision()) {
1494    HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1,
1495        startSpecifier, specifierLen);
1496  }
1497
1498  // Check each flag does not conflict with any other component.
1499  if (!FS.hasValidLeadingZeros())
1500    HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen);
1501  if (!FS.hasValidPlusPrefix())
1502    HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen);
1503  if (!FS.hasValidSpacePrefix())
1504    HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen);
1505  if (!FS.hasValidAlternativeForm())
1506    HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen);
1507  if (!FS.hasValidLeftJustified())
1508    HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen);
1509
1510  // Check that flags are not ignored by another flag
1511  if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+'
1512    HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(),
1513        startSpecifier, specifierLen);
1514  if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-'
1515    HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(),
1516            startSpecifier, specifierLen);
1517
1518  // Check the length modifier is valid with the given conversion specifier.
1519  const LengthModifier &LM = FS.getLengthModifier();
1520  if (!FS.hasValidLengthModifier())
1521    S.Diag(getLocationOfByte(LM.getStart()),
1522        diag::warn_format_nonsensical_length)
1523      << LM.toString() << CS.toString()
1524      << getSpecifierRange(startSpecifier, specifierLen)
1525      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1526          LM.getLength()));
1527
1528  // Are we using '%n'?
1529  if (CS.getKind() == ConversionSpecifier::nArg) {
1530    // Issue a warning about this being a possible security issue.
1531    S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back)
1532      << getSpecifierRange(startSpecifier, specifierLen);
1533    // Continue checking the other format specifiers.
1534    return true;
1535  }
1536
1537  // The remaining checks depend on the data arguments.
1538  if (HasVAListArg)
1539    return true;
1540
1541  if (argIndex >= NumDataArgs) {
1542    if (FS.usesPositionalArg())  {
1543      S.Diag(getLocationOfByte(CS.getStart()),
1544             diag::warn_printf_positional_arg_exceeds_data_args)
1545        << (argIndex+1) << NumDataArgs
1546        << getSpecifierRange(startSpecifier, specifierLen);
1547    }
1548    else {
1549      S.Diag(getLocationOfByte(CS.getStart()),
1550             diag::warn_printf_insufficient_data_args)
1551        << getSpecifierRange(startSpecifier, specifierLen);
1552    }
1553
1554    // Don't do any more checking.
1555    return false;
1556  }
1557
1558  // Now type check the data expression that matches the
1559  // format specifier.
1560  const Expr *Ex = getDataArg(argIndex);
1561  const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context);
1562  if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) {
1563    // Check if we didn't match because of an implicit cast from a 'char'
1564    // or 'short' to an 'int'.  This is done because printf is a varargs
1565    // function.
1566    if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex))
1567      if (ICE->getType() == S.Context.IntTy)
1568        if (ATR.matchesType(S.Context, ICE->getSubExpr()->getType()))
1569          return true;
1570
1571    // We may be able to offer a FixItHint if it is a supported type.
1572    PrintfSpecifier fixedFS = FS;
1573    bool success = fixedFS.fixType(Ex->getType());
1574
1575    if (success) {
1576      // Get the fix string from the fixed format specifier
1577      llvm::SmallString<128> buf;
1578      llvm::raw_svector_ostream os(buf);
1579      fixedFS.toString(os);
1580
1581      S.Diag(getLocationOfByte(CS.getStart()),
1582          diag::warn_printf_conversion_argument_type_mismatch)
1583        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1584        << getSpecifierRange(startSpecifier, specifierLen)
1585        << Ex->getSourceRange()
1586        << FixItHint::CreateReplacement(
1587            getSpecifierRange(startSpecifier, specifierLen),
1588            os.str());
1589    }
1590    else {
1591      S.Diag(getLocationOfByte(CS.getStart()),
1592             diag::warn_printf_conversion_argument_type_mismatch)
1593        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1594        << getSpecifierRange(startSpecifier, specifierLen)
1595        << Ex->getSourceRange();
1596    }
1597  }
1598
1599  return true;
1600}
1601
1602//===--- CHECK: Scanf format string checking ------------------------------===//
1603
1604namespace {
1605class CheckScanfHandler : public CheckFormatHandler {
1606public:
1607  CheckScanfHandler(Sema &s, const StringLiteral *fexpr,
1608                    const Expr *origFormatExpr, unsigned firstDataArg,
1609                    unsigned numDataArgs, bool isObjCLiteral,
1610                    const char *beg, bool hasVAListArg,
1611                    const CallExpr *theCall, unsigned formatIdx)
1612  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1613                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1614                       theCall, formatIdx) {}
1615
1616  bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS,
1617                            const char *startSpecifier,
1618                            unsigned specifierLen);
1619
1620  bool HandleInvalidScanfConversionSpecifier(
1621          const analyze_scanf::ScanfSpecifier &FS,
1622          const char *startSpecifier,
1623          unsigned specifierLen);
1624
1625  void HandleIncompleteScanList(const char *start, const char *end);
1626};
1627}
1628
1629void CheckScanfHandler::HandleIncompleteScanList(const char *start,
1630                                                 const char *end) {
1631  S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete)
1632    << getSpecifierRange(start, end - start);
1633}
1634
1635bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier(
1636                                        const analyze_scanf::ScanfSpecifier &FS,
1637                                        const char *startSpecifier,
1638                                        unsigned specifierLen) {
1639
1640  const analyze_scanf::ScanfConversionSpecifier &CS =
1641    FS.getConversionSpecifier();
1642
1643  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1644                                          getLocationOfByte(CS.getStart()),
1645                                          startSpecifier, specifierLen,
1646                                          CS.getStart(), CS.getLength());
1647}
1648
1649bool CheckScanfHandler::HandleScanfSpecifier(
1650                                       const analyze_scanf::ScanfSpecifier &FS,
1651                                       const char *startSpecifier,
1652                                       unsigned specifierLen) {
1653
1654  using namespace analyze_scanf;
1655  using namespace analyze_format_string;
1656
1657  const ScanfConversionSpecifier &CS = FS.getConversionSpecifier();
1658
1659  // Handle case where '%' and '*' don't consume an argument.  These shouldn't
1660  // be used to decide if we are using positional arguments consistently.
1661  if (FS.consumesDataArgument()) {
1662    if (atFirstArg) {
1663      atFirstArg = false;
1664      usesPositionalArgs = FS.usesPositionalArg();
1665    }
1666    else if (usesPositionalArgs != FS.usesPositionalArg()) {
1667      // Cannot mix-and-match positional and non-positional arguments.
1668      S.Diag(getLocationOfByte(CS.getStart()),
1669             diag::warn_format_mix_positional_nonpositional_args)
1670        << getSpecifierRange(startSpecifier, specifierLen);
1671      return false;
1672    }
1673  }
1674
1675  // Check if the field with is non-zero.
1676  const OptionalAmount &Amt = FS.getFieldWidth();
1677  if (Amt.getHowSpecified() == OptionalAmount::Constant) {
1678    if (Amt.getConstantAmount() == 0) {
1679      const CharSourceRange &R = getSpecifierRange(Amt.getStart(),
1680                                                   Amt.getConstantLength());
1681      S.Diag(getLocationOfByte(Amt.getStart()),
1682             diag::warn_scanf_nonzero_width)
1683        << R << FixItHint::CreateRemoval(R);
1684    }
1685  }
1686
1687  if (!FS.consumesDataArgument()) {
1688    // FIXME: Technically specifying a precision or field width here
1689    // makes no sense.  Worth issuing a warning at some point.
1690    return true;
1691  }
1692
1693  // Consume the argument.
1694  unsigned argIndex = FS.getArgIndex();
1695  if (argIndex < NumDataArgs) {
1696      // The check to see if the argIndex is valid will come later.
1697      // We set the bit here because we may exit early from this
1698      // function if we encounter some other error.
1699    CoveredArgs.set(argIndex);
1700  }
1701
1702  // Check the length modifier is valid with the given conversion specifier.
1703  const LengthModifier &LM = FS.getLengthModifier();
1704  if (!FS.hasValidLengthModifier()) {
1705    S.Diag(getLocationOfByte(LM.getStart()),
1706           diag::warn_format_nonsensical_length)
1707      << LM.toString() << CS.toString()
1708      << getSpecifierRange(startSpecifier, specifierLen)
1709      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1710                                                    LM.getLength()));
1711  }
1712
1713  // The remaining checks depend on the data arguments.
1714  if (HasVAListArg)
1715    return true;
1716
1717  if (argIndex >= NumDataArgs) {
1718    if (FS.usesPositionalArg())  {
1719      S.Diag(getLocationOfByte(CS.getStart()),
1720             diag::warn_printf_positional_arg_exceeds_data_args)
1721      << (argIndex+1) << NumDataArgs
1722      << getSpecifierRange(startSpecifier, specifierLen);
1723    }
1724    else {
1725      S.Diag(getLocationOfByte(CS.getStart()),
1726             diag::warn_printf_insufficient_data_args)
1727      << getSpecifierRange(startSpecifier, specifierLen);
1728    }
1729
1730    // Don't do any more checking.
1731    return false;
1732  }
1733
1734  // FIXME: Check that the argument type matches the format specifier.
1735
1736  return true;
1737}
1738
1739void Sema::CheckFormatString(const StringLiteral *FExpr,
1740                             const Expr *OrigFormatExpr,
1741                             const CallExpr *TheCall, bool HasVAListArg,
1742                             unsigned format_idx, unsigned firstDataArg,
1743                             bool isPrintf) {
1744
1745  // CHECK: is the format string a wide literal?
1746  if (FExpr->isWide()) {
1747    Diag(FExpr->getLocStart(),
1748         diag::warn_format_string_is_wide_literal)
1749    << OrigFormatExpr->getSourceRange();
1750    return;
1751  }
1752
1753  // Str - The format string.  NOTE: this is NOT null-terminated!
1754  const char *Str = FExpr->getStrData();
1755
1756  // CHECK: empty format string?
1757  unsigned StrLen = FExpr->getByteLength();
1758
1759  if (StrLen == 0) {
1760    Diag(FExpr->getLocStart(), diag::warn_empty_format_string)
1761    << OrigFormatExpr->getSourceRange();
1762    return;
1763  }
1764
1765  if (isPrintf) {
1766    CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1767                         TheCall->getNumArgs() - firstDataArg,
1768                         isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1769                         HasVAListArg, TheCall, format_idx);
1770
1771    if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen))
1772      H.DoneProcessing();
1773  }
1774  else {
1775    CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1776                        TheCall->getNumArgs() - firstDataArg,
1777                        isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1778                        HasVAListArg, TheCall, format_idx);
1779
1780    if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen))
1781      H.DoneProcessing();
1782  }
1783}
1784
1785//===--- CHECK: Return Address of Stack Variable --------------------------===//
1786
1787static DeclRefExpr* EvalVal(Expr *E);
1788static DeclRefExpr* EvalAddr(Expr* E);
1789
1790/// CheckReturnStackAddr - Check if a return statement returns the address
1791///   of a stack variable.
1792void
1793Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
1794                           SourceLocation ReturnLoc) {
1795
1796  // Perform checking for returned stack addresses.
1797  if (lhsType->isPointerType() || lhsType->isBlockPointerType()) {
1798    if (DeclRefExpr *DR = EvalAddr(RetValExp))
1799      Diag(DR->getLocStart(), diag::warn_ret_stack_addr)
1800       << DR->getDecl()->getDeclName() << RetValExp->getSourceRange();
1801
1802    // Skip over implicit cast expressions when checking for block expressions.
1803    RetValExp = RetValExp->IgnoreParenCasts();
1804
1805    if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp))
1806      if (C->hasBlockDeclRefExprs())
1807        Diag(C->getLocStart(), diag::err_ret_local_block)
1808          << C->getSourceRange();
1809
1810    if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp))
1811      Diag(ALE->getLocStart(), diag::warn_ret_addr_label)
1812        << ALE->getSourceRange();
1813
1814  } else if (lhsType->isReferenceType()) {
1815    // Perform checking for stack values returned by reference.
1816    // Check for a reference to the stack
1817    if (DeclRefExpr *DR = EvalVal(RetValExp))
1818      Diag(DR->getLocStart(), diag::warn_ret_stack_ref)
1819        << DR->getDecl()->getDeclName() << RetValExp->getSourceRange();
1820  }
1821}
1822
1823/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
1824///  check if the expression in a return statement evaluates to an address
1825///  to a location on the stack.  The recursion is used to traverse the
1826///  AST of the return expression, with recursion backtracking when we
1827///  encounter a subexpression that (1) clearly does not lead to the address
1828///  of a stack variable or (2) is something we cannot determine leads to
1829///  the address of a stack variable based on such local checking.
1830///
1831///  EvalAddr processes expressions that are pointers that are used as
1832///  references (and not L-values).  EvalVal handles all other values.
1833///  At the base case of the recursion is a check for a DeclRefExpr* in
1834///  the refers to a stack variable.
1835///
1836///  This implementation handles:
1837///
1838///   * pointer-to-pointer casts
1839///   * implicit conversions from array references to pointers
1840///   * taking the address of fields
1841///   * arbitrary interplay between "&" and "*" operators
1842///   * pointer arithmetic from an address of a stack variable
1843///   * taking the address of an array element where the array is on the stack
1844static DeclRefExpr* EvalAddr(Expr *E) {
1845  // We should only be called for evaluating pointer expressions.
1846  assert((E->getType()->isAnyPointerType() ||
1847          E->getType()->isBlockPointerType() ||
1848          E->getType()->isObjCQualifiedIdType()) &&
1849         "EvalAddr only works on pointers");
1850
1851  // Our "symbolic interpreter" is just a dispatch off the currently
1852  // viewed AST node.  We then recursively traverse the AST by calling
1853  // EvalAddr and EvalVal appropriately.
1854  switch (E->getStmtClass()) {
1855  case Stmt::ParenExprClass:
1856    // Ignore parentheses.
1857    return EvalAddr(cast<ParenExpr>(E)->getSubExpr());
1858
1859  case Stmt::UnaryOperatorClass: {
1860    // The only unary operator that make sense to handle here
1861    // is AddrOf.  All others don't make sense as pointers.
1862    UnaryOperator *U = cast<UnaryOperator>(E);
1863
1864    if (U->getOpcode() == UnaryOperator::AddrOf)
1865      return EvalVal(U->getSubExpr());
1866    else
1867      return NULL;
1868  }
1869
1870  case Stmt::BinaryOperatorClass: {
1871    // Handle pointer arithmetic.  All other binary operators are not valid
1872    // in this context.
1873    BinaryOperator *B = cast<BinaryOperator>(E);
1874    BinaryOperator::Opcode op = B->getOpcode();
1875
1876    if (op != BinaryOperator::Add && op != BinaryOperator::Sub)
1877      return NULL;
1878
1879    Expr *Base = B->getLHS();
1880
1881    // Determine which argument is the real pointer base.  It could be
1882    // the RHS argument instead of the LHS.
1883    if (!Base->getType()->isPointerType()) Base = B->getRHS();
1884
1885    assert (Base->getType()->isPointerType());
1886    return EvalAddr(Base);
1887  }
1888
1889  // For conditional operators we need to see if either the LHS or RHS are
1890  // valid DeclRefExpr*s.  If one of them is valid, we return it.
1891  case Stmt::ConditionalOperatorClass: {
1892    ConditionalOperator *C = cast<ConditionalOperator>(E);
1893
1894    // Handle the GNU extension for missing LHS.
1895    if (Expr *lhsExpr = C->getLHS())
1896      if (DeclRefExpr* LHS = EvalAddr(lhsExpr))
1897        return LHS;
1898
1899     return EvalAddr(C->getRHS());
1900  }
1901
1902  // For casts, we need to handle conversions from arrays to
1903  // pointer values, and pointer-to-pointer conversions.
1904  case Stmt::ImplicitCastExprClass:
1905  case Stmt::CStyleCastExprClass:
1906  case Stmt::CXXFunctionalCastExprClass: {
1907    Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
1908    QualType T = SubExpr->getType();
1909
1910    if (SubExpr->getType()->isPointerType() ||
1911        SubExpr->getType()->isBlockPointerType() ||
1912        SubExpr->getType()->isObjCQualifiedIdType())
1913      return EvalAddr(SubExpr);
1914    else if (T->isArrayType())
1915      return EvalVal(SubExpr);
1916    else
1917      return 0;
1918  }
1919
1920  // C++ casts.  For dynamic casts, static casts, and const casts, we
1921  // are always converting from a pointer-to-pointer, so we just blow
1922  // through the cast.  In the case the dynamic cast doesn't fail (and
1923  // return NULL), we take the conservative route and report cases
1924  // where we return the address of a stack variable.  For Reinterpre
1925  // FIXME: The comment about is wrong; we're not always converting
1926  // from pointer to pointer. I'm guessing that this code should also
1927  // handle references to objects.
1928  case Stmt::CXXStaticCastExprClass:
1929  case Stmt::CXXDynamicCastExprClass:
1930  case Stmt::CXXConstCastExprClass:
1931  case Stmt::CXXReinterpretCastExprClass: {
1932      Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr();
1933      if (S->getType()->isPointerType() || S->getType()->isBlockPointerType())
1934        return EvalAddr(S);
1935      else
1936        return NULL;
1937  }
1938
1939  // Everything else: we simply don't reason about them.
1940  default:
1941    return NULL;
1942  }
1943}
1944
1945
1946///  EvalVal - This function is complements EvalAddr in the mutual recursion.
1947///   See the comments for EvalAddr for more details.
1948static DeclRefExpr* EvalVal(Expr *E) {
1949
1950  // We should only be called for evaluating non-pointer expressions, or
1951  // expressions with a pointer type that are not used as references but instead
1952  // are l-values (e.g., DeclRefExpr with a pointer type).
1953
1954  // Our "symbolic interpreter" is just a dispatch off the currently
1955  // viewed AST node.  We then recursively traverse the AST by calling
1956  // EvalAddr and EvalVal appropriately.
1957  switch (E->getStmtClass()) {
1958  case Stmt::DeclRefExprClass: {
1959    // DeclRefExpr: the base case.  When we hit a DeclRefExpr we are looking
1960    //  at code that refers to a variable's name.  We check if it has local
1961    //  storage within the function, and if so, return the expression.
1962    DeclRefExpr *DR = cast<DeclRefExpr>(E);
1963
1964    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
1965      if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR;
1966
1967    return NULL;
1968  }
1969
1970  case Stmt::ParenExprClass:
1971    // Ignore parentheses.
1972    return EvalVal(cast<ParenExpr>(E)->getSubExpr());
1973
1974  case Stmt::UnaryOperatorClass: {
1975    // The only unary operator that make sense to handle here
1976    // is Deref.  All others don't resolve to a "name."  This includes
1977    // handling all sorts of rvalues passed to a unary operator.
1978    UnaryOperator *U = cast<UnaryOperator>(E);
1979
1980    if (U->getOpcode() == UnaryOperator::Deref)
1981      return EvalAddr(U->getSubExpr());
1982
1983    return NULL;
1984  }
1985
1986  case Stmt::ArraySubscriptExprClass: {
1987    // Array subscripts are potential references to data on the stack.  We
1988    // retrieve the DeclRefExpr* for the array variable if it indeed
1989    // has local storage.
1990    return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase());
1991  }
1992
1993  case Stmt::ConditionalOperatorClass: {
1994    // For conditional operators we need to see if either the LHS or RHS are
1995    // non-NULL DeclRefExpr's.  If one is non-NULL, we return it.
1996    ConditionalOperator *C = cast<ConditionalOperator>(E);
1997
1998    // Handle the GNU extension for missing LHS.
1999    if (Expr *lhsExpr = C->getLHS())
2000      if (DeclRefExpr *LHS = EvalVal(lhsExpr))
2001        return LHS;
2002
2003    return EvalVal(C->getRHS());
2004  }
2005
2006  // Accesses to members are potential references to data on the stack.
2007  case Stmt::MemberExprClass: {
2008    MemberExpr *M = cast<MemberExpr>(E);
2009
2010    // Check for indirect access.  We only want direct field accesses.
2011    if (!M->isArrow())
2012      return EvalVal(M->getBase());
2013    else
2014      return NULL;
2015  }
2016
2017  // Everything else: we simply don't reason about them.
2018  default:
2019    return NULL;
2020  }
2021}
2022
2023//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
2024
2025/// Check for comparisons of floating point operands using != and ==.
2026/// Issue a warning if these are no self-comparisons, as they are not likely
2027/// to do what the programmer intended.
2028void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) {
2029  bool EmitWarning = true;
2030
2031  Expr* LeftExprSansParen = lex->IgnoreParens();
2032  Expr* RightExprSansParen = rex->IgnoreParens();
2033
2034  // Special case: check for x == x (which is OK).
2035  // Do not emit warnings for such cases.
2036  if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
2037    if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
2038      if (DRL->getDecl() == DRR->getDecl())
2039        EmitWarning = false;
2040
2041
2042  // Special case: check for comparisons against literals that can be exactly
2043  //  represented by APFloat.  In such cases, do not emit a warning.  This
2044  //  is a heuristic: often comparison against such literals are used to
2045  //  detect if a value in a variable has not changed.  This clearly can
2046  //  lead to false negatives.
2047  if (EmitWarning) {
2048    if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
2049      if (FLL->isExact())
2050        EmitWarning = false;
2051    } else
2052      if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){
2053        if (FLR->isExact())
2054          EmitWarning = false;
2055    }
2056  }
2057
2058  // Check for comparisons with builtin types.
2059  if (EmitWarning)
2060    if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
2061      if (CL->isBuiltinCall(Context))
2062        EmitWarning = false;
2063
2064  if (EmitWarning)
2065    if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
2066      if (CR->isBuiltinCall(Context))
2067        EmitWarning = false;
2068
2069  // Emit the diagnostic.
2070  if (EmitWarning)
2071    Diag(loc, diag::warn_floatingpoint_eq)
2072      << lex->getSourceRange() << rex->getSourceRange();
2073}
2074
2075//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===//
2076//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===//
2077
2078namespace {
2079
2080/// Structure recording the 'active' range of an integer-valued
2081/// expression.
2082struct IntRange {
2083  /// The number of bits active in the int.
2084  unsigned Width;
2085
2086  /// True if the int is known not to have negative values.
2087  bool NonNegative;
2088
2089  IntRange() {}
2090  IntRange(unsigned Width, bool NonNegative)
2091    : Width(Width), NonNegative(NonNegative)
2092  {}
2093
2094  // Returns the range of the bool type.
2095  static IntRange forBoolType() {
2096    return IntRange(1, true);
2097  }
2098
2099  // Returns the range of an integral type.
2100  static IntRange forType(ASTContext &C, QualType T) {
2101    return forCanonicalType(C, T->getCanonicalTypeInternal().getTypePtr());
2102  }
2103
2104  // Returns the range of an integeral type based on its canonical
2105  // representation.
2106  static IntRange forCanonicalType(ASTContext &C, const Type *T) {
2107    assert(T->isCanonicalUnqualified());
2108
2109    if (const VectorType *VT = dyn_cast<VectorType>(T))
2110      T = VT->getElementType().getTypePtr();
2111    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2112      T = CT->getElementType().getTypePtr();
2113
2114    if (const EnumType *ET = dyn_cast<EnumType>(T)) {
2115      EnumDecl *Enum = ET->getDecl();
2116      unsigned NumPositive = Enum->getNumPositiveBits();
2117      unsigned NumNegative = Enum->getNumNegativeBits();
2118
2119      return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0);
2120    }
2121
2122    const BuiltinType *BT = cast<BuiltinType>(T);
2123    assert(BT->isInteger());
2124
2125    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2126  }
2127
2128  // Returns the supremum of two ranges: i.e. their conservative merge.
2129  static IntRange join(IntRange L, IntRange R) {
2130    return IntRange(std::max(L.Width, R.Width),
2131                    L.NonNegative && R.NonNegative);
2132  }
2133
2134  // Returns the infinum of two ranges: i.e. their aggressive merge.
2135  static IntRange meet(IntRange L, IntRange R) {
2136    return IntRange(std::min(L.Width, R.Width),
2137                    L.NonNegative || R.NonNegative);
2138  }
2139};
2140
2141IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) {
2142  if (value.isSigned() && value.isNegative())
2143    return IntRange(value.getMinSignedBits(), false);
2144
2145  if (value.getBitWidth() > MaxWidth)
2146    value.trunc(MaxWidth);
2147
2148  // isNonNegative() just checks the sign bit without considering
2149  // signedness.
2150  return IntRange(value.getActiveBits(), true);
2151}
2152
2153IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty,
2154                       unsigned MaxWidth) {
2155  if (result.isInt())
2156    return GetValueRange(C, result.getInt(), MaxWidth);
2157
2158  if (result.isVector()) {
2159    IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth);
2160    for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) {
2161      IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth);
2162      R = IntRange::join(R, El);
2163    }
2164    return R;
2165  }
2166
2167  if (result.isComplexInt()) {
2168    IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth);
2169    IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth);
2170    return IntRange::join(R, I);
2171  }
2172
2173  // This can happen with lossless casts to intptr_t of "based" lvalues.
2174  // Assume it might use arbitrary bits.
2175  // FIXME: The only reason we need to pass the type in here is to get
2176  // the sign right on this one case.  It would be nice if APValue
2177  // preserved this.
2178  assert(result.isLValue());
2179  return IntRange(MaxWidth, Ty->isUnsignedIntegerType());
2180}
2181
2182/// Pseudo-evaluate the given integer expression, estimating the
2183/// range of values it might take.
2184///
2185/// \param MaxWidth - the width to which the value will be truncated
2186IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) {
2187  E = E->IgnoreParens();
2188
2189  // Try a full evaluation first.
2190  Expr::EvalResult result;
2191  if (E->Evaluate(result, C))
2192    return GetValueRange(C, result.Val, E->getType(), MaxWidth);
2193
2194  // I think we only want to look through implicit casts here; if the
2195  // user has an explicit widening cast, we should treat the value as
2196  // being of the new, wider type.
2197  if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) {
2198    if (CE->getCastKind() == CastExpr::CK_NoOp)
2199      return GetExprRange(C, CE->getSubExpr(), MaxWidth);
2200
2201    IntRange OutputTypeRange = IntRange::forType(C, CE->getType());
2202
2203    bool isIntegerCast = (CE->getCastKind() == CastExpr::CK_IntegralCast);
2204    if (!isIntegerCast && CE->getCastKind() == CastExpr::CK_Unknown)
2205      isIntegerCast = CE->getSubExpr()->getType()->isIntegerType();
2206
2207    // Assume that non-integer casts can span the full range of the type.
2208    if (!isIntegerCast)
2209      return OutputTypeRange;
2210
2211    IntRange SubRange
2212      = GetExprRange(C, CE->getSubExpr(),
2213                     std::min(MaxWidth, OutputTypeRange.Width));
2214
2215    // Bail out if the subexpr's range is as wide as the cast type.
2216    if (SubRange.Width >= OutputTypeRange.Width)
2217      return OutputTypeRange;
2218
2219    // Otherwise, we take the smaller width, and we're non-negative if
2220    // either the output type or the subexpr is.
2221    return IntRange(SubRange.Width,
2222                    SubRange.NonNegative || OutputTypeRange.NonNegative);
2223  }
2224
2225  if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
2226    // If we can fold the condition, just take that operand.
2227    bool CondResult;
2228    if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C))
2229      return GetExprRange(C, CondResult ? CO->getTrueExpr()
2230                                        : CO->getFalseExpr(),
2231                          MaxWidth);
2232
2233    // Otherwise, conservatively merge.
2234    IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth);
2235    IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth);
2236    return IntRange::join(L, R);
2237  }
2238
2239  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
2240    switch (BO->getOpcode()) {
2241
2242    // Boolean-valued operations are single-bit and positive.
2243    case BinaryOperator::LAnd:
2244    case BinaryOperator::LOr:
2245    case BinaryOperator::LT:
2246    case BinaryOperator::GT:
2247    case BinaryOperator::LE:
2248    case BinaryOperator::GE:
2249    case BinaryOperator::EQ:
2250    case BinaryOperator::NE:
2251      return IntRange::forBoolType();
2252
2253    // The type of these compound assignments is the type of the LHS,
2254    // so the RHS is not necessarily an integer.
2255    case BinaryOperator::MulAssign:
2256    case BinaryOperator::DivAssign:
2257    case BinaryOperator::RemAssign:
2258    case BinaryOperator::AddAssign:
2259    case BinaryOperator::SubAssign:
2260      return IntRange::forType(C, E->getType());
2261
2262    // Operations with opaque sources are black-listed.
2263    case BinaryOperator::PtrMemD:
2264    case BinaryOperator::PtrMemI:
2265      return IntRange::forType(C, E->getType());
2266
2267    // Bitwise-and uses the *infinum* of the two source ranges.
2268    case BinaryOperator::And:
2269    case BinaryOperator::AndAssign:
2270      return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth),
2271                            GetExprRange(C, BO->getRHS(), MaxWidth));
2272
2273    // Left shift gets black-listed based on a judgement call.
2274    case BinaryOperator::Shl:
2275      // ...except that we want to treat '1 << (blah)' as logically
2276      // positive.  It's an important idiom.
2277      if (IntegerLiteral *I
2278            = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) {
2279        if (I->getValue() == 1) {
2280          IntRange R = IntRange::forType(C, E->getType());
2281          return IntRange(R.Width, /*NonNegative*/ true);
2282        }
2283      }
2284      // fallthrough
2285
2286    case BinaryOperator::ShlAssign:
2287      return IntRange::forType(C, E->getType());
2288
2289    // Right shift by a constant can narrow its left argument.
2290    case BinaryOperator::Shr:
2291    case BinaryOperator::ShrAssign: {
2292      IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2293
2294      // If the shift amount is a positive constant, drop the width by
2295      // that much.
2296      llvm::APSInt shift;
2297      if (BO->getRHS()->isIntegerConstantExpr(shift, C) &&
2298          shift.isNonNegative()) {
2299        unsigned zext = shift.getZExtValue();
2300        if (zext >= L.Width)
2301          L.Width = (L.NonNegative ? 0 : 1);
2302        else
2303          L.Width -= zext;
2304      }
2305
2306      return L;
2307    }
2308
2309    // Comma acts as its right operand.
2310    case BinaryOperator::Comma:
2311      return GetExprRange(C, BO->getRHS(), MaxWidth);
2312
2313    // Black-list pointer subtractions.
2314    case BinaryOperator::Sub:
2315      if (BO->getLHS()->getType()->isPointerType())
2316        return IntRange::forType(C, E->getType());
2317      // fallthrough
2318
2319    default:
2320      break;
2321    }
2322
2323    // Treat every other operator as if it were closed on the
2324    // narrowest type that encompasses both operands.
2325    IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2326    IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth);
2327    return IntRange::join(L, R);
2328  }
2329
2330  if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
2331    switch (UO->getOpcode()) {
2332    // Boolean-valued operations are white-listed.
2333    case UnaryOperator::LNot:
2334      return IntRange::forBoolType();
2335
2336    // Operations with opaque sources are black-listed.
2337    case UnaryOperator::Deref:
2338    case UnaryOperator::AddrOf: // should be impossible
2339    case UnaryOperator::OffsetOf:
2340      return IntRange::forType(C, E->getType());
2341
2342    default:
2343      return GetExprRange(C, UO->getSubExpr(), MaxWidth);
2344    }
2345  }
2346
2347  if (dyn_cast<OffsetOfExpr>(E)) {
2348    IntRange::forType(C, E->getType());
2349  }
2350
2351  FieldDecl *BitField = E->getBitField();
2352  if (BitField) {
2353    llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C);
2354    unsigned BitWidth = BitWidthAP.getZExtValue();
2355
2356    return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType());
2357  }
2358
2359  return IntRange::forType(C, E->getType());
2360}
2361
2362IntRange GetExprRange(ASTContext &C, Expr *E) {
2363  return GetExprRange(C, E, C.getIntWidth(E->getType()));
2364}
2365
2366/// Checks whether the given value, which currently has the given
2367/// source semantics, has the same value when coerced through the
2368/// target semantics.
2369bool IsSameFloatAfterCast(const llvm::APFloat &value,
2370                          const llvm::fltSemantics &Src,
2371                          const llvm::fltSemantics &Tgt) {
2372  llvm::APFloat truncated = value;
2373
2374  bool ignored;
2375  truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored);
2376  truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored);
2377
2378  return truncated.bitwiseIsEqual(value);
2379}
2380
2381/// Checks whether the given value, which currently has the given
2382/// source semantics, has the same value when coerced through the
2383/// target semantics.
2384///
2385/// The value might be a vector of floats (or a complex number).
2386bool IsSameFloatAfterCast(const APValue &value,
2387                          const llvm::fltSemantics &Src,
2388                          const llvm::fltSemantics &Tgt) {
2389  if (value.isFloat())
2390    return IsSameFloatAfterCast(value.getFloat(), Src, Tgt);
2391
2392  if (value.isVector()) {
2393    for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i)
2394      if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt))
2395        return false;
2396    return true;
2397  }
2398
2399  assert(value.isComplexFloat());
2400  return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) &&
2401          IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt));
2402}
2403
2404void AnalyzeImplicitConversions(Sema &S, Expr *E);
2405
2406bool IsZero(Sema &S, Expr *E) {
2407  llvm::APSInt Value;
2408  return E->isIntegerConstantExpr(Value, S.Context) && Value == 0;
2409}
2410
2411void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) {
2412  BinaryOperator::Opcode op = E->getOpcode();
2413  if (op == BinaryOperator::LT && IsZero(S, E->getRHS())) {
2414    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2415      << "< 0" << "false"
2416      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2417  } else if (op == BinaryOperator::GE && IsZero(S, E->getRHS())) {
2418    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2419      << ">= 0" << "true"
2420      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2421  } else if (op == BinaryOperator::GT && IsZero(S, E->getLHS())) {
2422    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
2423      << "0 >" << "false"
2424      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2425  } else if (op == BinaryOperator::LE && IsZero(S, E->getLHS())) {
2426    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
2427      << "0 <=" << "true"
2428      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2429  }
2430}
2431
2432/// Analyze the operands of the given comparison.  Implements the
2433/// fallback case from AnalyzeComparison.
2434void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) {
2435  AnalyzeImplicitConversions(S, E->getLHS());
2436  AnalyzeImplicitConversions(S, E->getRHS());
2437}
2438
2439/// \brief Implements -Wsign-compare.
2440///
2441/// \param lex the left-hand expression
2442/// \param rex the right-hand expression
2443/// \param OpLoc the location of the joining operator
2444/// \param BinOpc binary opcode or 0
2445void AnalyzeComparison(Sema &S, BinaryOperator *E) {
2446  // The type the comparison is being performed in.
2447  QualType T = E->getLHS()->getType();
2448  assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType())
2449         && "comparison with mismatched types");
2450
2451  // We don't do anything special if this isn't an unsigned integral
2452  // comparison:  we're only interested in integral comparisons, and
2453  // signed comparisons only happen in cases we don't care to warn about.
2454  if (!T->hasUnsignedIntegerRepresentation())
2455    return AnalyzeImpConvsInComparison(S, E);
2456
2457  Expr *lex = E->getLHS()->IgnoreParenImpCasts();
2458  Expr *rex = E->getRHS()->IgnoreParenImpCasts();
2459
2460  // Check to see if one of the (unmodified) operands is of different
2461  // signedness.
2462  Expr *signedOperand, *unsignedOperand;
2463  if (lex->getType()->hasSignedIntegerRepresentation()) {
2464    assert(!rex->getType()->hasSignedIntegerRepresentation() &&
2465           "unsigned comparison between two signed integer expressions?");
2466    signedOperand = lex;
2467    unsignedOperand = rex;
2468  } else if (rex->getType()->hasSignedIntegerRepresentation()) {
2469    signedOperand = rex;
2470    unsignedOperand = lex;
2471  } else {
2472    CheckTrivialUnsignedComparison(S, E);
2473    return AnalyzeImpConvsInComparison(S, E);
2474  }
2475
2476  // Otherwise, calculate the effective range of the signed operand.
2477  IntRange signedRange = GetExprRange(S.Context, signedOperand);
2478
2479  // Go ahead and analyze implicit conversions in the operands.  Note
2480  // that we skip the implicit conversions on both sides.
2481  AnalyzeImplicitConversions(S, lex);
2482  AnalyzeImplicitConversions(S, rex);
2483
2484  // If the signed range is non-negative, -Wsign-compare won't fire,
2485  // but we should still check for comparisons which are always true
2486  // or false.
2487  if (signedRange.NonNegative)
2488    return CheckTrivialUnsignedComparison(S, E);
2489
2490  // For (in)equality comparisons, if the unsigned operand is a
2491  // constant which cannot collide with a overflowed signed operand,
2492  // then reinterpreting the signed operand as unsigned will not
2493  // change the result of the comparison.
2494  if (E->isEqualityOp()) {
2495    unsigned comparisonWidth = S.Context.getIntWidth(T);
2496    IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand);
2497
2498    // We should never be unable to prove that the unsigned operand is
2499    // non-negative.
2500    assert(unsignedRange.NonNegative && "unsigned range includes negative?");
2501
2502    if (unsignedRange.Width < comparisonWidth)
2503      return;
2504  }
2505
2506  S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison)
2507    << lex->getType() << rex->getType()
2508    << lex->getSourceRange() << rex->getSourceRange();
2509}
2510
2511/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
2512void DiagnoseImpCast(Sema &S, Expr *E, QualType T, unsigned diag) {
2513  S.Diag(E->getExprLoc(), diag) << E->getType() << T << E->getSourceRange();
2514}
2515
2516void CheckImplicitConversion(Sema &S, Expr *E, QualType T,
2517                             bool *ICContext = 0) {
2518  if (E->isTypeDependent() || E->isValueDependent()) return;
2519
2520  const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr();
2521  const Type *Target = S.Context.getCanonicalType(T).getTypePtr();
2522  if (Source == Target) return;
2523  if (Target->isDependentType()) return;
2524
2525  // Never diagnose implicit casts to bool.
2526  if (Target->isSpecificBuiltinType(BuiltinType::Bool))
2527    return;
2528
2529  // Strip vector types.
2530  if (isa<VectorType>(Source)) {
2531    if (!isa<VectorType>(Target))
2532      return DiagnoseImpCast(S, E, T, diag::warn_impcast_vector_scalar);
2533
2534    Source = cast<VectorType>(Source)->getElementType().getTypePtr();
2535    Target = cast<VectorType>(Target)->getElementType().getTypePtr();
2536  }
2537
2538  // Strip complex types.
2539  if (isa<ComplexType>(Source)) {
2540    if (!isa<ComplexType>(Target))
2541      return DiagnoseImpCast(S, E, T, diag::warn_impcast_complex_scalar);
2542
2543    Source = cast<ComplexType>(Source)->getElementType().getTypePtr();
2544    Target = cast<ComplexType>(Target)->getElementType().getTypePtr();
2545  }
2546
2547  const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source);
2548  const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target);
2549
2550  // If the source is floating point...
2551  if (SourceBT && SourceBT->isFloatingPoint()) {
2552    // ...and the target is floating point...
2553    if (TargetBT && TargetBT->isFloatingPoint()) {
2554      // ...then warn if we're dropping FP rank.
2555
2556      // Builtin FP kinds are ordered by increasing FP rank.
2557      if (SourceBT->getKind() > TargetBT->getKind()) {
2558        // Don't warn about float constants that are precisely
2559        // representable in the target type.
2560        Expr::EvalResult result;
2561        if (E->Evaluate(result, S.Context)) {
2562          // Value might be a float, a float vector, or a float complex.
2563          if (IsSameFloatAfterCast(result.Val,
2564                   S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)),
2565                   S.Context.getFloatTypeSemantics(QualType(SourceBT, 0))))
2566            return;
2567        }
2568
2569        DiagnoseImpCast(S, E, T, diag::warn_impcast_float_precision);
2570      }
2571      return;
2572    }
2573
2574    // If the target is integral, always warn.
2575    if ((TargetBT && TargetBT->isInteger()))
2576      // TODO: don't warn for integer values?
2577      DiagnoseImpCast(S, E, T, diag::warn_impcast_float_integer);
2578
2579    return;
2580  }
2581
2582  if (!Source->isIntegerType() || !Target->isIntegerType())
2583    return;
2584
2585  IntRange SourceRange = GetExprRange(S.Context, E);
2586  IntRange TargetRange = IntRange::forCanonicalType(S.Context, Target);
2587
2588  if (SourceRange.Width > TargetRange.Width) {
2589    // People want to build with -Wshorten-64-to-32 and not -Wconversion
2590    // and by god we'll let them.
2591    if (SourceRange.Width == 64 && TargetRange.Width == 32)
2592      return DiagnoseImpCast(S, E, T, diag::warn_impcast_integer_64_32);
2593    return DiagnoseImpCast(S, E, T, diag::warn_impcast_integer_precision);
2594  }
2595
2596  if ((TargetRange.NonNegative && !SourceRange.NonNegative) ||
2597      (!TargetRange.NonNegative && SourceRange.NonNegative &&
2598       SourceRange.Width == TargetRange.Width)) {
2599    unsigned DiagID = diag::warn_impcast_integer_sign;
2600
2601    // Traditionally, gcc has warned about this under -Wsign-compare.
2602    // We also want to warn about it in -Wconversion.
2603    // So if -Wconversion is off, use a completely identical diagnostic
2604    // in the sign-compare group.
2605    // The conditional-checking code will
2606    if (ICContext) {
2607      DiagID = diag::warn_impcast_integer_sign_conditional;
2608      *ICContext = true;
2609    }
2610
2611    return DiagnoseImpCast(S, E, T, DiagID);
2612  }
2613
2614  return;
2615}
2616
2617void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T);
2618
2619void CheckConditionalOperand(Sema &S, Expr *E, QualType T,
2620                             bool &ICContext) {
2621  E = E->IgnoreParenImpCasts();
2622
2623  if (isa<ConditionalOperator>(E))
2624    return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T);
2625
2626  AnalyzeImplicitConversions(S, E);
2627  if (E->getType() != T)
2628    return CheckImplicitConversion(S, E, T, &ICContext);
2629  return;
2630}
2631
2632void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) {
2633  AnalyzeImplicitConversions(S, E->getCond());
2634
2635  bool Suspicious = false;
2636  CheckConditionalOperand(S, E->getTrueExpr(), T, Suspicious);
2637  CheckConditionalOperand(S, E->getFalseExpr(), T, Suspicious);
2638
2639  // If -Wconversion would have warned about either of the candidates
2640  // for a signedness conversion to the context type...
2641  if (!Suspicious) return;
2642
2643  // ...but it's currently ignored...
2644  if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional))
2645    return;
2646
2647  // ...and -Wsign-compare isn't...
2648  if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional))
2649    return;
2650
2651  // ...then check whether it would have warned about either of the
2652  // candidates for a signedness conversion to the condition type.
2653  if (E->getType() != T) {
2654    Suspicious = false;
2655    CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(),
2656                            E->getType(), &Suspicious);
2657    if (!Suspicious)
2658      CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(),
2659                              E->getType(), &Suspicious);
2660    if (!Suspicious)
2661      return;
2662  }
2663
2664  // If so, emit a diagnostic under -Wsign-compare.
2665  Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts();
2666  Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts();
2667  S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional)
2668    << lex->getType() << rex->getType()
2669    << lex->getSourceRange() << rex->getSourceRange();
2670}
2671
2672/// AnalyzeImplicitConversions - Find and report any interesting
2673/// implicit conversions in the given expression.  There are a couple
2674/// of competing diagnostics here, -Wconversion and -Wsign-compare.
2675void AnalyzeImplicitConversions(Sema &S, Expr *OrigE) {
2676  QualType T = OrigE->getType();
2677  Expr *E = OrigE->IgnoreParenImpCasts();
2678
2679  // For conditional operators, we analyze the arguments as if they
2680  // were being fed directly into the output.
2681  if (isa<ConditionalOperator>(E)) {
2682    ConditionalOperator *CO = cast<ConditionalOperator>(E);
2683    CheckConditionalOperator(S, CO, T);
2684    return;
2685  }
2686
2687  // Go ahead and check any implicit conversions we might have skipped.
2688  // The non-canonical typecheck is just an optimization;
2689  // CheckImplicitConversion will filter out dead implicit conversions.
2690  if (E->getType() != T)
2691    CheckImplicitConversion(S, E, T);
2692
2693  // Now continue drilling into this expression.
2694
2695  // Skip past explicit casts.
2696  if (isa<ExplicitCastExpr>(E)) {
2697    E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts();
2698    return AnalyzeImplicitConversions(S, E);
2699  }
2700
2701  // Do a somewhat different check with comparison operators.
2702  if (isa<BinaryOperator>(E) && cast<BinaryOperator>(E)->isComparisonOp())
2703    return AnalyzeComparison(S, cast<BinaryOperator>(E));
2704
2705  // These break the otherwise-useful invariant below.  Fortunately,
2706  // we don't really need to recurse into them, because any internal
2707  // expressions should have been analyzed already when they were
2708  // built into statements.
2709  if (isa<StmtExpr>(E)) return;
2710
2711  // Don't descend into unevaluated contexts.
2712  if (isa<SizeOfAlignOfExpr>(E)) return;
2713
2714  // Now just recurse over the expression's children.
2715  for (Stmt::child_iterator I = E->child_begin(), IE = E->child_end();
2716         I != IE; ++I)
2717    AnalyzeImplicitConversions(S, cast<Expr>(*I));
2718}
2719
2720} // end anonymous namespace
2721
2722/// Diagnoses "dangerous" implicit conversions within the given
2723/// expression (which is a full expression).  Implements -Wconversion
2724/// and -Wsign-compare.
2725void Sema::CheckImplicitConversions(Expr *E) {
2726  // Don't diagnose in unevaluated contexts.
2727  if (ExprEvalContexts.back().Context == Sema::Unevaluated)
2728    return;
2729
2730  // Don't diagnose for value- or type-dependent expressions.
2731  if (E->isTypeDependent() || E->isValueDependent())
2732    return;
2733
2734  AnalyzeImplicitConversions(*this, E);
2735}
2736
2737/// CheckParmsForFunctionDef - Check that the parameters of the given
2738/// function are appropriate for the definition of a function. This
2739/// takes care of any checks that cannot be performed on the
2740/// declaration itself, e.g., that the types of each of the function
2741/// parameters are complete.
2742bool Sema::CheckParmsForFunctionDef(FunctionDecl *FD) {
2743  bool HasInvalidParm = false;
2744  for (unsigned p = 0, NumParams = FD->getNumParams(); p < NumParams; ++p) {
2745    ParmVarDecl *Param = FD->getParamDecl(p);
2746
2747    // C99 6.7.5.3p4: the parameters in a parameter type list in a
2748    // function declarator that is part of a function definition of
2749    // that function shall not have incomplete type.
2750    //
2751    // This is also C++ [dcl.fct]p6.
2752    if (!Param->isInvalidDecl() &&
2753        RequireCompleteType(Param->getLocation(), Param->getType(),
2754                               diag::err_typecheck_decl_incomplete_type)) {
2755      Param->setInvalidDecl();
2756      HasInvalidParm = true;
2757    }
2758
2759    // C99 6.9.1p5: If the declarator includes a parameter type list, the
2760    // declaration of each parameter shall include an identifier.
2761    if (Param->getIdentifier() == 0 &&
2762        !Param->isImplicit() &&
2763        !getLangOptions().CPlusPlus)
2764      Diag(Param->getLocation(), diag::err_parameter_name_omitted);
2765
2766    // C99 6.7.5.3p12:
2767    //   If the function declarator is not part of a definition of that
2768    //   function, parameters may have incomplete type and may use the [*]
2769    //   notation in their sequences of declarator specifiers to specify
2770    //   variable length array types.
2771    QualType PType = Param->getOriginalType();
2772    if (const ArrayType *AT = Context.getAsArrayType(PType)) {
2773      if (AT->getSizeModifier() == ArrayType::Star) {
2774        // FIXME: This diagnosic should point the the '[*]' if source-location
2775        // information is added for it.
2776        Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
2777      }
2778    }
2779  }
2780
2781  return HasInvalidParm;
2782}
2783