SemaChecking.cpp revision b075f1de60c17bd36ec1eaaf9e560224f8dbf7e4
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  // We need to figure out which concrete builtin this maps onto.  For example,
423  // __sync_fetch_and_add with a 2 byte object turns into
424  // __sync_fetch_and_add_2.
425#define BUILTIN_ROW(x) \
426  { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
427    Builtin::BI##x##_8, Builtin::BI##x##_16 }
428
429  static const unsigned BuiltinIndices[][5] = {
430    BUILTIN_ROW(__sync_fetch_and_add),
431    BUILTIN_ROW(__sync_fetch_and_sub),
432    BUILTIN_ROW(__sync_fetch_and_or),
433    BUILTIN_ROW(__sync_fetch_and_and),
434    BUILTIN_ROW(__sync_fetch_and_xor),
435
436    BUILTIN_ROW(__sync_add_and_fetch),
437    BUILTIN_ROW(__sync_sub_and_fetch),
438    BUILTIN_ROW(__sync_and_and_fetch),
439    BUILTIN_ROW(__sync_or_and_fetch),
440    BUILTIN_ROW(__sync_xor_and_fetch),
441
442    BUILTIN_ROW(__sync_val_compare_and_swap),
443    BUILTIN_ROW(__sync_bool_compare_and_swap),
444    BUILTIN_ROW(__sync_lock_test_and_set),
445    BUILTIN_ROW(__sync_lock_release)
446  };
447#undef BUILTIN_ROW
448
449  // Determine the index of the size.
450  unsigned SizeIndex;
451  switch (Context.getTypeSizeInChars(ValType).getQuantity()) {
452  case 1: SizeIndex = 0; break;
453  case 2: SizeIndex = 1; break;
454  case 4: SizeIndex = 2; break;
455  case 8: SizeIndex = 3; break;
456  case 16: SizeIndex = 4; break;
457  default:
458    Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
459      << FirstArg->getType() << FirstArg->getSourceRange();
460    return ExprError();
461  }
462
463  // Each of these builtins has one pointer argument, followed by some number of
464  // values (0, 1 or 2) followed by a potentially empty varags list of stuff
465  // that we ignore.  Find out which row of BuiltinIndices to read from as well
466  // as the number of fixed args.
467  unsigned BuiltinID = FDecl->getBuiltinID();
468  unsigned BuiltinIndex, NumFixed = 1;
469  switch (BuiltinID) {
470  default: assert(0 && "Unknown overloaded atomic builtin!");
471  case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break;
472  case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break;
473  case Builtin::BI__sync_fetch_and_or:  BuiltinIndex = 2; break;
474  case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break;
475  case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break;
476
477  case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break;
478  case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break;
479  case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break;
480  case Builtin::BI__sync_or_and_fetch:  BuiltinIndex = 8; break;
481  case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break;
482
483  case Builtin::BI__sync_val_compare_and_swap:
484    BuiltinIndex = 10;
485    NumFixed = 2;
486    break;
487  case Builtin::BI__sync_bool_compare_and_swap:
488    BuiltinIndex = 11;
489    NumFixed = 2;
490    break;
491  case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break;
492  case Builtin::BI__sync_lock_release:
493    BuiltinIndex = 13;
494    NumFixed = 0;
495    break;
496  }
497
498  // Now that we know how many fixed arguments we expect, first check that we
499  // have at least that many.
500  if (TheCall->getNumArgs() < 1+NumFixed) {
501    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
502      << 0 << 1+NumFixed << TheCall->getNumArgs()
503      << TheCall->getCallee()->getSourceRange();
504    return ExprError();
505  }
506
507  // Get the decl for the concrete builtin from this, we can tell what the
508  // concrete integer type we should convert to is.
509  unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
510  const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
511  IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName);
512  FunctionDecl *NewBuiltinDecl =
513    cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID,
514                                           TUScope, false, DRE->getLocStart()));
515
516  // The first argument is by definition correct, we use it's type as the type
517  // of the entire operation. Walk the remaining arguments promoting them to
518  // the deduced value type.
519  for (unsigned i = 0; i != NumFixed; ++i) {
520    Expr *Arg = TheCall->getArg(i+1);
521
522    // If the argument is an implicit cast, then there was a promotion due to
523    // "...", just remove it now.
524    if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg)) {
525      Arg = ICE->getSubExpr();
526      ICE->setSubExpr(0);
527      ICE->Destroy(Context);
528      TheCall->setArg(i+1, Arg);
529    }
530
531    // GCC does an implicit conversion to the pointer or integer ValType.  This
532    // can fail in some cases (1i -> int**), check for this error case now.
533    CastExpr::CastKind Kind = CastExpr::CK_Unknown;
534    CXXBaseSpecifierArray BasePath;
535    if (CheckCastTypes(Arg->getSourceRange(), ValType, Arg, Kind, BasePath))
536      return ExprError();
537
538    // Okay, we have something that *can* be converted to the right type.  Check
539    // to see if there is a potentially weird extension going on here.  This can
540    // happen when you do an atomic operation on something like an char* and
541    // pass in 42.  The 42 gets converted to char.  This is even more strange
542    // for things like 45.123 -> char, etc.
543    // FIXME: Do this check.
544    ImpCastExprToType(Arg, ValType, Kind);
545    TheCall->setArg(i+1, Arg);
546  }
547
548  // Switch the DeclRefExpr to refer to the new decl.
549  DRE->setDecl(NewBuiltinDecl);
550  DRE->setType(NewBuiltinDecl->getType());
551
552  // Set the callee in the CallExpr.
553  // FIXME: This leaks the original parens and implicit casts.
554  Expr *PromotedCall = DRE;
555  UsualUnaryConversions(PromotedCall);
556  TheCall->setCallee(PromotedCall);
557
558  // Change the result type of the call to match the original value type. This
559  // is arbitrary, but the codegen for these builtins ins design to handle it
560  // gracefully.
561  TheCall->setType(ValType);
562
563  return move(TheCallResult);
564}
565
566
567/// CheckObjCString - Checks that the argument to the builtin
568/// CFString constructor is correct
569/// FIXME: GCC currently emits the following warning:
570/// "warning: input conversion stopped due to an input byte that does not
571///           belong to the input codeset UTF-8"
572/// Note: It might also make sense to do the UTF-16 conversion here (would
573/// simplify the backend).
574bool Sema::CheckObjCString(Expr *Arg) {
575  Arg = Arg->IgnoreParenCasts();
576  StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
577
578  if (!Literal || Literal->isWide()) {
579    Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
580      << Arg->getSourceRange();
581    return true;
582  }
583
584  const char *Data = Literal->getStrData();
585  unsigned Length = Literal->getByteLength();
586
587  for (unsigned i = 0; i < Length; ++i) {
588    if (!Data[i]) {
589      Diag(getLocationOfStringLiteralByte(Literal, i),
590           diag::warn_cfstring_literal_contains_nul_character)
591        << Arg->getSourceRange();
592      break;
593    }
594  }
595
596  return false;
597}
598
599/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
600/// Emit an error and return true on failure, return false on success.
601bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
602  Expr *Fn = TheCall->getCallee();
603  if (TheCall->getNumArgs() > 2) {
604    Diag(TheCall->getArg(2)->getLocStart(),
605         diag::err_typecheck_call_too_many_args)
606      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
607      << Fn->getSourceRange()
608      << SourceRange(TheCall->getArg(2)->getLocStart(),
609                     (*(TheCall->arg_end()-1))->getLocEnd());
610    return true;
611  }
612
613  if (TheCall->getNumArgs() < 2) {
614    return Diag(TheCall->getLocEnd(),
615      diag::err_typecheck_call_too_few_args_at_least)
616      << 0 /*function call*/ << 2 << TheCall->getNumArgs();
617  }
618
619  // Determine whether the current function is variadic or not.
620  BlockScopeInfo *CurBlock = getCurBlock();
621  bool isVariadic;
622  if (CurBlock)
623    isVariadic = CurBlock->TheDecl->isVariadic();
624  else if (FunctionDecl *FD = getCurFunctionDecl())
625    isVariadic = FD->isVariadic();
626  else
627    isVariadic = getCurMethodDecl()->isVariadic();
628
629  if (!isVariadic) {
630    Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
631    return true;
632  }
633
634  // Verify that the second argument to the builtin is the last argument of the
635  // current function or method.
636  bool SecondArgIsLastNamedArgument = false;
637  const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
638
639  if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
640    if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
641      // FIXME: This isn't correct for methods (results in bogus warning).
642      // Get the last formal in the current function.
643      const ParmVarDecl *LastArg;
644      if (CurBlock)
645        LastArg = *(CurBlock->TheDecl->param_end()-1);
646      else if (FunctionDecl *FD = getCurFunctionDecl())
647        LastArg = *(FD->param_end()-1);
648      else
649        LastArg = *(getCurMethodDecl()->param_end()-1);
650      SecondArgIsLastNamedArgument = PV == LastArg;
651    }
652  }
653
654  if (!SecondArgIsLastNamedArgument)
655    Diag(TheCall->getArg(1)->getLocStart(),
656         diag::warn_second_parameter_of_va_start_not_last_named_argument);
657  return false;
658}
659
660/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
661/// friends.  This is declared to take (...), so we have to check everything.
662bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
663  if (TheCall->getNumArgs() < 2)
664    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
665      << 0 << 2 << TheCall->getNumArgs()/*function call*/;
666  if (TheCall->getNumArgs() > 2)
667    return Diag(TheCall->getArg(2)->getLocStart(),
668                diag::err_typecheck_call_too_many_args)
669      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
670      << SourceRange(TheCall->getArg(2)->getLocStart(),
671                     (*(TheCall->arg_end()-1))->getLocEnd());
672
673  Expr *OrigArg0 = TheCall->getArg(0);
674  Expr *OrigArg1 = TheCall->getArg(1);
675
676  // Do standard promotions between the two arguments, returning their common
677  // type.
678  QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
679
680  // Make sure any conversions are pushed back into the call; this is
681  // type safe since unordered compare builtins are declared as "_Bool
682  // foo(...)".
683  TheCall->setArg(0, OrigArg0);
684  TheCall->setArg(1, OrigArg1);
685
686  if (OrigArg0->isTypeDependent() || OrigArg1->isTypeDependent())
687    return false;
688
689  // If the common type isn't a real floating type, then the arguments were
690  // invalid for this operation.
691  if (!Res->isRealFloatingType())
692    return Diag(OrigArg0->getLocStart(),
693                diag::err_typecheck_call_invalid_ordered_compare)
694      << OrigArg0->getType() << OrigArg1->getType()
695      << SourceRange(OrigArg0->getLocStart(), OrigArg1->getLocEnd());
696
697  return false;
698}
699
700/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like
701/// __builtin_isnan and friends.  This is declared to take (...), so we have
702/// to check everything. We expect the last argument to be a floating point
703/// value.
704bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) {
705  if (TheCall->getNumArgs() < NumArgs)
706    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
707      << 0 << NumArgs << TheCall->getNumArgs()/*function call*/;
708  if (TheCall->getNumArgs() > NumArgs)
709    return Diag(TheCall->getArg(NumArgs)->getLocStart(),
710                diag::err_typecheck_call_too_many_args)
711      << 0 /*function call*/ << NumArgs << TheCall->getNumArgs()
712      << SourceRange(TheCall->getArg(NumArgs)->getLocStart(),
713                     (*(TheCall->arg_end()-1))->getLocEnd());
714
715  Expr *OrigArg = TheCall->getArg(NumArgs-1);
716
717  if (OrigArg->isTypeDependent())
718    return false;
719
720  // This operation requires a non-_Complex floating-point number.
721  if (!OrigArg->getType()->isRealFloatingType())
722    return Diag(OrigArg->getLocStart(),
723                diag::err_typecheck_call_invalid_unary_fp)
724      << OrigArg->getType() << OrigArg->getSourceRange();
725
726  // If this is an implicit conversion from float -> double, remove it.
727  if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) {
728    Expr *CastArg = Cast->getSubExpr();
729    if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) {
730      assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) &&
731             "promotion from float to double is the only expected cast here");
732      Cast->setSubExpr(0);
733      Cast->Destroy(Context);
734      TheCall->setArg(NumArgs-1, CastArg);
735      OrigArg = CastArg;
736    }
737  }
738
739  return false;
740}
741
742/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
743// This is declared to take (...), so we have to check everything.
744Action::OwningExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
745  if (TheCall->getNumArgs() < 2)
746    return ExprError(Diag(TheCall->getLocEnd(),
747                          diag::err_typecheck_call_too_few_args_at_least)
748      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
749      << TheCall->getSourceRange());
750
751  // Determine which of the following types of shufflevector we're checking:
752  // 1) unary, vector mask: (lhs, mask)
753  // 2) binary, vector mask: (lhs, rhs, mask)
754  // 3) binary, scalar mask: (lhs, rhs, index, ..., index)
755  QualType resType = TheCall->getArg(0)->getType();
756  unsigned numElements = 0;
757
758  if (!TheCall->getArg(0)->isTypeDependent() &&
759      !TheCall->getArg(1)->isTypeDependent()) {
760    QualType LHSType = TheCall->getArg(0)->getType();
761    QualType RHSType = TheCall->getArg(1)->getType();
762
763    if (!LHSType->isVectorType() || !RHSType->isVectorType()) {
764      Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
765        << SourceRange(TheCall->getArg(0)->getLocStart(),
766                       TheCall->getArg(1)->getLocEnd());
767      return ExprError();
768    }
769
770    numElements = LHSType->getAs<VectorType>()->getNumElements();
771    unsigned numResElements = TheCall->getNumArgs() - 2;
772
773    // Check to see if we have a call with 2 vector arguments, the unary shuffle
774    // with mask.  If so, verify that RHS is an integer vector type with the
775    // same number of elts as lhs.
776    if (TheCall->getNumArgs() == 2) {
777      if (!RHSType->isIntegerType() ||
778          RHSType->getAs<VectorType>()->getNumElements() != numElements)
779        Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
780          << SourceRange(TheCall->getArg(1)->getLocStart(),
781                         TheCall->getArg(1)->getLocEnd());
782      numResElements = numElements;
783    }
784    else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) {
785      Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
786        << SourceRange(TheCall->getArg(0)->getLocStart(),
787                       TheCall->getArg(1)->getLocEnd());
788      return ExprError();
789    } else if (numElements != numResElements) {
790      QualType eltType = LHSType->getAs<VectorType>()->getElementType();
791      resType = Context.getVectorType(eltType, numResElements,
792                                      VectorType::NotAltiVec);
793    }
794  }
795
796  for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
797    if (TheCall->getArg(i)->isTypeDependent() ||
798        TheCall->getArg(i)->isValueDependent())
799      continue;
800
801    llvm::APSInt Result(32);
802    if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
803      return ExprError(Diag(TheCall->getLocStart(),
804                  diag::err_shufflevector_nonconstant_argument)
805                << TheCall->getArg(i)->getSourceRange());
806
807    if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
808      return ExprError(Diag(TheCall->getLocStart(),
809                  diag::err_shufflevector_argument_too_large)
810               << TheCall->getArg(i)->getSourceRange());
811  }
812
813  llvm::SmallVector<Expr*, 32> exprs;
814
815  for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
816    exprs.push_back(TheCall->getArg(i));
817    TheCall->setArg(i, 0);
818  }
819
820  return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(),
821                                            exprs.size(), resType,
822                                            TheCall->getCallee()->getLocStart(),
823                                            TheCall->getRParenLoc()));
824}
825
826/// SemaBuiltinPrefetch - Handle __builtin_prefetch.
827// This is declared to take (const void*, ...) and can take two
828// optional constant int args.
829bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
830  unsigned NumArgs = TheCall->getNumArgs();
831
832  if (NumArgs > 3)
833    return Diag(TheCall->getLocEnd(),
834             diag::err_typecheck_call_too_many_args_at_most)
835             << 0 /*function call*/ << 3 << NumArgs
836             << TheCall->getSourceRange();
837
838  // Argument 0 is checked for us and the remaining arguments must be
839  // constant integers.
840  for (unsigned i = 1; i != NumArgs; ++i) {
841    Expr *Arg = TheCall->getArg(i);
842
843    llvm::APSInt Result;
844    if (SemaBuiltinConstantArg(TheCall, i, Result))
845      return true;
846
847    // FIXME: gcc issues a warning and rewrites these to 0. These
848    // seems especially odd for the third argument since the default
849    // is 3.
850    if (i == 1) {
851      if (Result.getLimitedValue() > 1)
852        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
853             << "0" << "1" << Arg->getSourceRange();
854    } else {
855      if (Result.getLimitedValue() > 3)
856        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
857            << "0" << "3" << Arg->getSourceRange();
858    }
859  }
860
861  return false;
862}
863
864/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr
865/// TheCall is a constant expression.
866bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,
867                                  llvm::APSInt &Result) {
868  Expr *Arg = TheCall->getArg(ArgNum);
869  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
870  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
871
872  if (Arg->isTypeDependent() || Arg->isValueDependent()) return false;
873
874  if (!Arg->isIntegerConstantExpr(Result, Context))
875    return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type)
876                << FDecl->getDeclName() <<  Arg->getSourceRange();
877
878  return false;
879}
880
881/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
882/// int type). This simply type checks that type is one of the defined
883/// constants (0-3).
884// For compatability check 0-3, llvm only handles 0 and 2.
885bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
886  llvm::APSInt Result;
887
888  // Check constant-ness first.
889  if (SemaBuiltinConstantArg(TheCall, 1, Result))
890    return true;
891
892  Expr *Arg = TheCall->getArg(1);
893  if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
894    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
895             << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
896  }
897
898  return false;
899}
900
901/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
902/// This checks that val is a constant 1.
903bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
904  Expr *Arg = TheCall->getArg(1);
905  llvm::APSInt Result;
906
907  // TODO: This is less than ideal. Overload this to take a value.
908  if (SemaBuiltinConstantArg(TheCall, 1, Result))
909    return true;
910
911  if (Result != 1)
912    return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
913             << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
914
915  return false;
916}
917
918// Handle i > 1 ? "x" : "y", recursivelly
919bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall,
920                                  bool HasVAListArg,
921                                  unsigned format_idx, unsigned firstDataArg,
922                                  bool isPrintf) {
923
924  if (E->isTypeDependent() || E->isValueDependent())
925    return false;
926
927  switch (E->getStmtClass()) {
928  case Stmt::ConditionalOperatorClass: {
929    const ConditionalOperator *C = cast<ConditionalOperator>(E);
930    return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg,
931                                  format_idx, firstDataArg, isPrintf)
932        && SemaCheckStringLiteral(C->getRHS(), TheCall, HasVAListArg,
933                                  format_idx, firstDataArg, isPrintf);
934  }
935
936  case Stmt::ImplicitCastExprClass: {
937    const ImplicitCastExpr *Expr = cast<ImplicitCastExpr>(E);
938    return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg,
939                                  format_idx, firstDataArg, isPrintf);
940  }
941
942  case Stmt::ParenExprClass: {
943    const ParenExpr *Expr = cast<ParenExpr>(E);
944    return SemaCheckStringLiteral(Expr->getSubExpr(), TheCall, HasVAListArg,
945                                  format_idx, firstDataArg, isPrintf);
946  }
947
948  case Stmt::DeclRefExprClass: {
949    const DeclRefExpr *DR = cast<DeclRefExpr>(E);
950
951    // As an exception, do not flag errors for variables binding to
952    // const string literals.
953    if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
954      bool isConstant = false;
955      QualType T = DR->getType();
956
957      if (const ArrayType *AT = Context.getAsArrayType(T)) {
958        isConstant = AT->getElementType().isConstant(Context);
959      } else if (const PointerType *PT = T->getAs<PointerType>()) {
960        isConstant = T.isConstant(Context) &&
961                     PT->getPointeeType().isConstant(Context);
962      }
963
964      if (isConstant) {
965        if (const Expr *Init = VD->getAnyInitializer())
966          return SemaCheckStringLiteral(Init, TheCall,
967                                        HasVAListArg, format_idx, firstDataArg,
968                                        isPrintf);
969      }
970
971      // For vprintf* functions (i.e., HasVAListArg==true), we add a
972      // special check to see if the format string is a function parameter
973      // of the function calling the printf function.  If the function
974      // has an attribute indicating it is a printf-like function, then we
975      // should suppress warnings concerning non-literals being used in a call
976      // to a vprintf function.  For example:
977      //
978      // void
979      // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
980      //      va_list ap;
981      //      va_start(ap, fmt);
982      //      vprintf(fmt, ap);  // Do NOT emit a warning about "fmt".
983      //      ...
984      //
985      //
986      //  FIXME: We don't have full attribute support yet, so just check to see
987      //    if the argument is a DeclRefExpr that references a parameter.  We'll
988      //    add proper support for checking the attribute later.
989      if (HasVAListArg)
990        if (isa<ParmVarDecl>(VD))
991          return true;
992    }
993
994    return false;
995  }
996
997  case Stmt::CallExprClass: {
998    const CallExpr *CE = cast<CallExpr>(E);
999    if (const ImplicitCastExpr *ICE
1000          = dyn_cast<ImplicitCastExpr>(CE->getCallee())) {
1001      if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) {
1002        if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) {
1003          if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) {
1004            unsigned ArgIndex = FA->getFormatIdx();
1005            const Expr *Arg = CE->getArg(ArgIndex - 1);
1006
1007            return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg,
1008                                          format_idx, firstDataArg, isPrintf);
1009          }
1010        }
1011      }
1012    }
1013
1014    return false;
1015  }
1016  case Stmt::ObjCStringLiteralClass:
1017  case Stmt::StringLiteralClass: {
1018    const StringLiteral *StrE = NULL;
1019
1020    if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
1021      StrE = ObjCFExpr->getString();
1022    else
1023      StrE = cast<StringLiteral>(E);
1024
1025    if (StrE) {
1026      CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx,
1027                        firstDataArg, isPrintf);
1028      return true;
1029    }
1030
1031    return false;
1032  }
1033
1034  default:
1035    return false;
1036  }
1037}
1038
1039void
1040Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
1041                            const CallExpr *TheCall) {
1042  for (NonNullAttr::iterator i = NonNull->begin(), e = NonNull->end();
1043       i != e; ++i) {
1044    const Expr *ArgExpr = TheCall->getArg(*i);
1045    if (ArgExpr->isNullPointerConstant(Context,
1046                                       Expr::NPC_ValueDependentIsNotNull))
1047      Diag(TheCall->getCallee()->getLocStart(), diag::warn_null_arg)
1048        << ArgExpr->getSourceRange();
1049  }
1050}
1051
1052/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar
1053/// functions) for correct use of format strings.
1054void
1055Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg,
1056                                unsigned format_idx, unsigned firstDataArg,
1057                                bool isPrintf) {
1058
1059  const Expr *Fn = TheCall->getCallee();
1060
1061  // The way the format attribute works in GCC, the implicit this argument
1062  // of member functions is counted. However, it doesn't appear in our own
1063  // lists, so decrement format_idx in that case.
1064  if (isa<CXXMemberCallExpr>(TheCall)) {
1065    // Catch a format attribute mistakenly referring to the object argument.
1066    if (format_idx == 0)
1067      return;
1068    --format_idx;
1069    if(firstDataArg != 0)
1070      --firstDataArg;
1071  }
1072
1073  // CHECK: printf/scanf-like function is called with no format string.
1074  if (format_idx >= TheCall->getNumArgs()) {
1075    Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string)
1076      << Fn->getSourceRange();
1077    return;
1078  }
1079
1080  const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts();
1081
1082  // CHECK: format string is not a string literal.
1083  //
1084  // Dynamically generated format strings are difficult to
1085  // automatically vet at compile time.  Requiring that format strings
1086  // are string literals: (1) permits the checking of format strings by
1087  // the compiler and thereby (2) can practically remove the source of
1088  // many format string exploits.
1089
1090  // Format string can be either ObjC string (e.g. @"%d") or
1091  // C string (e.g. "%d")
1092  // ObjC string uses the same format specifiers as C string, so we can use
1093  // the same format string checking logic for both ObjC and C strings.
1094  if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx,
1095                             firstDataArg, isPrintf))
1096    return;  // Literal format string found, check done!
1097
1098  // If there are no arguments specified, warn with -Wformat-security, otherwise
1099  // warn only with -Wformat-nonliteral.
1100  if (TheCall->getNumArgs() == format_idx+1)
1101    Diag(TheCall->getArg(format_idx)->getLocStart(),
1102         diag::warn_format_nonliteral_noargs)
1103      << OrigFormatExpr->getSourceRange();
1104  else
1105    Diag(TheCall->getArg(format_idx)->getLocStart(),
1106         diag::warn_format_nonliteral)
1107           << OrigFormatExpr->getSourceRange();
1108}
1109
1110namespace {
1111class CheckFormatHandler : public analyze_format_string::FormatStringHandler {
1112protected:
1113  Sema &S;
1114  const StringLiteral *FExpr;
1115  const Expr *OrigFormatExpr;
1116  const unsigned FirstDataArg;
1117  const unsigned NumDataArgs;
1118  const bool IsObjCLiteral;
1119  const char *Beg; // Start of format string.
1120  const bool HasVAListArg;
1121  const CallExpr *TheCall;
1122  unsigned FormatIdx;
1123  llvm::BitVector CoveredArgs;
1124  bool usesPositionalArgs;
1125  bool atFirstArg;
1126public:
1127  CheckFormatHandler(Sema &s, const StringLiteral *fexpr,
1128                     const Expr *origFormatExpr, unsigned firstDataArg,
1129                     unsigned numDataArgs, bool isObjCLiteral,
1130                     const char *beg, bool hasVAListArg,
1131                     const CallExpr *theCall, unsigned formatIdx)
1132    : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr),
1133      FirstDataArg(firstDataArg),
1134      NumDataArgs(numDataArgs),
1135      IsObjCLiteral(isObjCLiteral), Beg(beg),
1136      HasVAListArg(hasVAListArg),
1137      TheCall(theCall), FormatIdx(formatIdx),
1138      usesPositionalArgs(false), atFirstArg(true) {
1139        CoveredArgs.resize(numDataArgs);
1140        CoveredArgs.reset();
1141      }
1142
1143  void DoneProcessing();
1144
1145  void HandleIncompleteSpecifier(const char *startSpecifier,
1146                                 unsigned specifierLen);
1147
1148  virtual void HandleInvalidPosition(const char *startSpecifier,
1149                                     unsigned specifierLen,
1150                                     analyze_format_string::PositionContext p);
1151
1152  virtual void HandleZeroPosition(const char *startPos, unsigned posLen);
1153
1154  void HandleNullChar(const char *nullCharacter);
1155
1156protected:
1157  SourceRange getFormatStringRange();
1158  CharSourceRange getSpecifierRange(const char *startSpecifier,
1159                                    unsigned specifierLen);
1160  SourceLocation getLocationOfByte(const char *x);
1161
1162  const Expr *getDataArg(unsigned i) const;
1163};
1164}
1165
1166SourceRange CheckFormatHandler::getFormatStringRange() {
1167  return OrigFormatExpr->getSourceRange();
1168}
1169
1170CharSourceRange CheckFormatHandler::
1171getSpecifierRange(const char *startSpecifier, unsigned specifierLen) {
1172  SourceLocation Start = getLocationOfByte(startSpecifier);
1173  SourceLocation End   = getLocationOfByte(startSpecifier + specifierLen - 1);
1174
1175  // Advance the end SourceLocation by one due to half-open ranges.
1176  End = End.getFileLocWithOffset(1);
1177
1178  return CharSourceRange::getCharRange(Start, End);
1179}
1180
1181SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) {
1182  return S.getLocationOfStringLiteralByte(FExpr, x - Beg);
1183}
1184
1185void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier,
1186                                                   unsigned specifierLen){
1187  SourceLocation Loc = getLocationOfByte(startSpecifier);
1188  S.Diag(Loc, diag::warn_printf_incomplete_specifier)
1189    << getSpecifierRange(startSpecifier, specifierLen);
1190}
1191
1192void
1193CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen,
1194                                     analyze_format_string::PositionContext p) {
1195  SourceLocation Loc = getLocationOfByte(startPos);
1196  S.Diag(Loc, diag::warn_format_invalid_positional_specifier)
1197    << (unsigned) p << getSpecifierRange(startPos, posLen);
1198}
1199
1200void CheckFormatHandler::HandleZeroPosition(const char *startPos,
1201                                            unsigned posLen) {
1202  SourceLocation Loc = getLocationOfByte(startPos);
1203  S.Diag(Loc, diag::warn_format_zero_positional_specifier)
1204    << getSpecifierRange(startPos, posLen);
1205}
1206
1207void CheckFormatHandler::HandleNullChar(const char *nullCharacter) {
1208  // The presence of a null character is likely an error.
1209  S.Diag(getLocationOfByte(nullCharacter),
1210         diag::warn_printf_format_string_contains_null_char)
1211    << getFormatStringRange();
1212}
1213
1214const Expr *CheckFormatHandler::getDataArg(unsigned i) const {
1215  return TheCall->getArg(FirstDataArg + i);
1216}
1217
1218void CheckFormatHandler::DoneProcessing() {
1219    // Does the number of data arguments exceed the number of
1220    // format conversions in the format string?
1221  if (!HasVAListArg) {
1222      // Find any arguments that weren't covered.
1223    CoveredArgs.flip();
1224    signed notCoveredArg = CoveredArgs.find_first();
1225    if (notCoveredArg >= 0) {
1226      assert((unsigned)notCoveredArg < NumDataArgs);
1227      S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(),
1228             diag::warn_printf_data_arg_not_used)
1229      << getFormatStringRange();
1230    }
1231  }
1232}
1233
1234//===--- CHECK: Printf format string checking ------------------------------===//
1235
1236namespace {
1237class CheckPrintfHandler : public CheckFormatHandler {
1238public:
1239  CheckPrintfHandler(Sema &s, const StringLiteral *fexpr,
1240                     const Expr *origFormatExpr, unsigned firstDataArg,
1241                     unsigned numDataArgs, bool isObjCLiteral,
1242                     const char *beg, bool hasVAListArg,
1243                     const CallExpr *theCall, unsigned formatIdx)
1244  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1245                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1246                       theCall, formatIdx) {}
1247
1248
1249  bool HandleInvalidPrintfConversionSpecifier(
1250                                      const analyze_printf::PrintfSpecifier &FS,
1251                                      const char *startSpecifier,
1252                                      unsigned specifierLen);
1253
1254  bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS,
1255                             const char *startSpecifier,
1256                             unsigned specifierLen);
1257
1258  bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k,
1259                    const char *startSpecifier, unsigned specifierLen);
1260  void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS,
1261                           const analyze_printf::OptionalAmount &Amt,
1262                           unsigned type,
1263                           const char *startSpecifier, unsigned specifierLen);
1264  void HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1265                  const analyze_printf::OptionalFlag &flag,
1266                  const char *startSpecifier, unsigned specifierLen);
1267  void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS,
1268                         const analyze_printf::OptionalFlag &ignoredFlag,
1269                         const analyze_printf::OptionalFlag &flag,
1270                         const char *startSpecifier, unsigned specifierLen);
1271};
1272}
1273
1274bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier(
1275                                      const analyze_printf::PrintfSpecifier &FS,
1276                                      const char *startSpecifier,
1277                                      unsigned specifierLen) {
1278
1279  unsigned argIndex = FS.getArgIndex();
1280  bool keepGoing = true;
1281  if (argIndex < NumDataArgs) {
1282      // Consider the argument coverered, even though the specifier doesn't
1283      // make sense.
1284    CoveredArgs.set(argIndex);
1285  }
1286  else {
1287    // If argIndex exceeds the number of data arguments we
1288    // don't issue a warning because that is just a cascade of warnings (and
1289    // they may have intended '%%' anyway). We don't want to continue processing
1290    // the format string after this point, however, as we will like just get
1291    // gibberish when trying to match arguments.
1292    keepGoing = false;
1293  }
1294
1295  const analyze_printf::ConversionSpecifier &CS =
1296  FS.getConversionSpecifier();
1297  SourceLocation Loc = getLocationOfByte(CS.getStart());
1298  S.Diag(Loc, diag::warn_printf_invalid_conversion)
1299  << llvm::StringRef(CS.getStart(), CS.getLength())
1300  << getSpecifierRange(startSpecifier, specifierLen);
1301
1302  return keepGoing;
1303}
1304
1305bool CheckPrintfHandler::HandleAmount(
1306                               const analyze_format_string::OptionalAmount &Amt,
1307                               unsigned k, const char *startSpecifier,
1308                               unsigned specifierLen) {
1309
1310  if (Amt.hasDataArgument()) {
1311    if (!HasVAListArg) {
1312      unsigned argIndex = Amt.getArgIndex();
1313      if (argIndex >= NumDataArgs) {
1314        S.Diag(getLocationOfByte(Amt.getStart()),
1315               diag::warn_printf_asterisk_missing_arg)
1316          << k << getSpecifierRange(startSpecifier, specifierLen);
1317        // Don't do any more checking.  We will just emit
1318        // spurious errors.
1319        return false;
1320      }
1321
1322      // Type check the data argument.  It should be an 'int'.
1323      // Although not in conformance with C99, we also allow the argument to be
1324      // an 'unsigned int' as that is a reasonably safe case.  GCC also
1325      // doesn't emit a warning for that case.
1326      CoveredArgs.set(argIndex);
1327      const Expr *Arg = getDataArg(argIndex);
1328      QualType T = Arg->getType();
1329
1330      const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context);
1331      assert(ATR.isValid());
1332
1333      if (!ATR.matchesType(S.Context, T)) {
1334        S.Diag(getLocationOfByte(Amt.getStart()),
1335               diag::warn_printf_asterisk_wrong_type)
1336          << k
1337          << ATR.getRepresentativeType(S.Context) << T
1338          << getSpecifierRange(startSpecifier, specifierLen)
1339          << Arg->getSourceRange();
1340        // Don't do any more checking.  We will just emit
1341        // spurious errors.
1342        return false;
1343      }
1344    }
1345  }
1346  return true;
1347}
1348
1349void CheckPrintfHandler::HandleInvalidAmount(
1350                                      const analyze_printf::PrintfSpecifier &FS,
1351                                      const analyze_printf::OptionalAmount &Amt,
1352                                      unsigned type,
1353                                      const char *startSpecifier,
1354                                      unsigned specifierLen) {
1355  const analyze_printf::ConversionSpecifier &CS = FS.getConversionSpecifier();
1356  switch (Amt.getHowSpecified()) {
1357  case analyze_printf::OptionalAmount::Constant:
1358    S.Diag(getLocationOfByte(Amt.getStart()),
1359        diag::warn_printf_nonsensical_optional_amount)
1360      << type
1361      << CS.toString()
1362      << getSpecifierRange(startSpecifier, specifierLen)
1363      << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(),
1364          Amt.getConstantLength()));
1365    break;
1366
1367  default:
1368    S.Diag(getLocationOfByte(Amt.getStart()),
1369        diag::warn_printf_nonsensical_optional_amount)
1370      << type
1371      << CS.toString()
1372      << getSpecifierRange(startSpecifier, specifierLen);
1373    break;
1374  }
1375}
1376
1377void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1378                                    const analyze_printf::OptionalFlag &flag,
1379                                    const char *startSpecifier,
1380                                    unsigned specifierLen) {
1381  // Warn about pointless flag with a fixit removal.
1382  const analyze_printf::ConversionSpecifier &CS = FS.getConversionSpecifier();
1383  S.Diag(getLocationOfByte(flag.getPosition()),
1384      diag::warn_printf_nonsensical_flag)
1385    << flag.toString() << CS.toString()
1386    << getSpecifierRange(startSpecifier, specifierLen)
1387    << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1));
1388}
1389
1390void CheckPrintfHandler::HandleIgnoredFlag(
1391                                const analyze_printf::PrintfSpecifier &FS,
1392                                const analyze_printf::OptionalFlag &ignoredFlag,
1393                                const analyze_printf::OptionalFlag &flag,
1394                                const char *startSpecifier,
1395                                unsigned specifierLen) {
1396  // Warn about ignored flag with a fixit removal.
1397  S.Diag(getLocationOfByte(ignoredFlag.getPosition()),
1398      diag::warn_printf_ignored_flag)
1399    << ignoredFlag.toString() << flag.toString()
1400    << getSpecifierRange(startSpecifier, specifierLen)
1401    << FixItHint::CreateRemoval(getSpecifierRange(
1402        ignoredFlag.getPosition(), 1));
1403}
1404
1405bool
1406CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
1407                                            &FS,
1408                                          const char *startSpecifier,
1409                                          unsigned specifierLen) {
1410
1411  using namespace analyze_printf;
1412  const ConversionSpecifier &CS = FS.getConversionSpecifier();
1413
1414  if (atFirstArg) {
1415    atFirstArg = false;
1416    usesPositionalArgs = FS.usesPositionalArg();
1417  }
1418  else if (usesPositionalArgs != FS.usesPositionalArg()) {
1419    // Cannot mix-and-match positional and non-positional arguments.
1420    S.Diag(getLocationOfByte(CS.getStart()),
1421           diag::warn_format_mix_positional_nonpositional_args)
1422      << getSpecifierRange(startSpecifier, specifierLen);
1423    return false;
1424  }
1425
1426  // First check if the field width, precision, and conversion specifier
1427  // have matching data arguments.
1428  if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0,
1429                    startSpecifier, specifierLen)) {
1430    return false;
1431  }
1432
1433  if (!HandleAmount(FS.getPrecision(), /* precision */ 1,
1434                    startSpecifier, specifierLen)) {
1435    return false;
1436  }
1437
1438  if (!CS.consumesDataArgument()) {
1439    // FIXME: Technically specifying a precision or field width here
1440    // makes no sense.  Worth issuing a warning at some point.
1441    return true;
1442  }
1443
1444  // Consume the argument.
1445  unsigned argIndex = FS.getArgIndex();
1446  if (argIndex < NumDataArgs) {
1447    // The check to see if the argIndex is valid will come later.
1448    // We set the bit here because we may exit early from this
1449    // function if we encounter some other error.
1450    CoveredArgs.set(argIndex);
1451  }
1452
1453  // Check for using an Objective-C specific conversion specifier
1454  // in a non-ObjC literal.
1455  if (!IsObjCLiteral && CS.isObjCArg()) {
1456    return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier,
1457                                                  specifierLen);
1458  }
1459
1460  // Check for invalid use of field width
1461  if (!FS.hasValidFieldWidth()) {
1462    HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0,
1463        startSpecifier, specifierLen);
1464  }
1465
1466  // Check for invalid use of precision
1467  if (!FS.hasValidPrecision()) {
1468    HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1,
1469        startSpecifier, specifierLen);
1470  }
1471
1472  // Check each flag does not conflict with any other component.
1473  if (!FS.hasValidLeadingZeros())
1474    HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen);
1475  if (!FS.hasValidPlusPrefix())
1476    HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen);
1477  if (!FS.hasValidSpacePrefix())
1478    HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen);
1479  if (!FS.hasValidAlternativeForm())
1480    HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen);
1481  if (!FS.hasValidLeftJustified())
1482    HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen);
1483
1484  // Check that flags are not ignored by another flag
1485  if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+'
1486    HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(),
1487        startSpecifier, specifierLen);
1488  if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-'
1489    HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(),
1490            startSpecifier, specifierLen);
1491
1492  // Check the length modifier is valid with the given conversion specifier.
1493  const LengthModifier &LM = FS.getLengthModifier();
1494  if (!FS.hasValidLengthModifier())
1495    S.Diag(getLocationOfByte(LM.getStart()),
1496        diag::warn_printf_nonsensical_length)
1497      << LM.toString() << CS.toString()
1498      << getSpecifierRange(startSpecifier, specifierLen)
1499      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1500          LM.getLength()));
1501
1502  // Are we using '%n'?
1503  if (CS.getKind() == ConversionSpecifier::OutIntPtrArg) {
1504    // Issue a warning about this being a possible security issue.
1505    S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back)
1506      << getSpecifierRange(startSpecifier, specifierLen);
1507    // Continue checking the other format specifiers.
1508    return true;
1509  }
1510
1511  // The remaining checks depend on the data arguments.
1512  if (HasVAListArg)
1513    return true;
1514
1515  if (argIndex >= NumDataArgs) {
1516    if (FS.usesPositionalArg())  {
1517      S.Diag(getLocationOfByte(CS.getStart()),
1518             diag::warn_printf_positional_arg_exceeds_data_args)
1519        << (argIndex+1) << NumDataArgs
1520        << getSpecifierRange(startSpecifier, specifierLen);
1521    }
1522    else {
1523      S.Diag(getLocationOfByte(CS.getStart()),
1524             diag::warn_printf_insufficient_data_args)
1525        << getSpecifierRange(startSpecifier, specifierLen);
1526    }
1527
1528    // Don't do any more checking.
1529    return false;
1530  }
1531
1532  // Now type check the data expression that matches the
1533  // format specifier.
1534  const Expr *Ex = getDataArg(argIndex);
1535  const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context);
1536  if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) {
1537    // Check if we didn't match because of an implicit cast from a 'char'
1538    // or 'short' to an 'int'.  This is done because printf is a varargs
1539    // function.
1540    if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex))
1541      if (ICE->getType() == S.Context.IntTy)
1542        if (ATR.matchesType(S.Context, ICE->getSubExpr()->getType()))
1543          return true;
1544
1545    // We may be able to offer a FixItHint if it is a supported type.
1546    PrintfSpecifier fixedFS = FS;
1547    bool success = fixedFS.fixType(Ex->getType());
1548
1549    if (success) {
1550      // Get the fix string from the fixed format specifier
1551      llvm::SmallString<128> buf;
1552      llvm::raw_svector_ostream os(buf);
1553      fixedFS.toString(os);
1554
1555      S.Diag(getLocationOfByte(CS.getStart()),
1556          diag::warn_printf_conversion_argument_type_mismatch)
1557        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1558        << getSpecifierRange(startSpecifier, specifierLen)
1559        << Ex->getSourceRange()
1560        << FixItHint::CreateReplacement(
1561            getSpecifierRange(startSpecifier, specifierLen),
1562            os.str());
1563    }
1564    else {
1565      S.Diag(getLocationOfByte(CS.getStart()),
1566             diag::warn_printf_conversion_argument_type_mismatch)
1567        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1568        << getSpecifierRange(startSpecifier, specifierLen)
1569        << Ex->getSourceRange();
1570    }
1571  }
1572
1573  return true;
1574}
1575
1576//===--- CHECK: Scanf format string checking ------------------------------===//
1577
1578namespace {
1579class CheckScanfHandler : public CheckFormatHandler {
1580public:
1581  CheckScanfHandler(Sema &s, const StringLiteral *fexpr,
1582                    const Expr *origFormatExpr, unsigned firstDataArg,
1583                    unsigned numDataArgs, bool isObjCLiteral,
1584                    const char *beg, bool hasVAListArg,
1585                    const CallExpr *theCall, unsigned formatIdx)
1586  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1587                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1588                       theCall, formatIdx) {}
1589
1590  bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS,
1591                            const char *startSpecifier,
1592                            unsigned specifierLen);
1593
1594  void HandleIncompleteScanList(const char *start, const char *end);
1595};
1596}
1597
1598void CheckScanfHandler::HandleIncompleteScanList(const char *start,
1599                                                 const char *end) {
1600  S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete)
1601    << getSpecifierRange(start, end - start);
1602}
1603
1604bool CheckScanfHandler::HandleScanfSpecifier(
1605                                       const analyze_scanf::ScanfSpecifier &FS,
1606                                       const char *startSpecifier,
1607                                       unsigned specifierLen) {
1608
1609  using namespace analyze_scanf;
1610  using namespace analyze_format_string;
1611
1612  const ConversionSpecifier &CS = FS.getConversionSpecifier();
1613
1614  // FIXME: Handle case where '%' and '*' don't consume an argument.
1615  // This needs to be done for the printf case as well.
1616  if (atFirstArg) {
1617    atFirstArg = false;
1618    usesPositionalArgs = FS.usesPositionalArg();
1619  }
1620  else if (usesPositionalArgs != FS.usesPositionalArg()) {
1621    // Cannot mix-and-match positional and non-positional arguments.
1622    S.Diag(getLocationOfByte(CS.getStart()),
1623           diag::warn_format_mix_positional_nonpositional_args)
1624      << getSpecifierRange(startSpecifier, specifierLen);
1625    return false;
1626  }
1627
1628  // Check if the field with is non-zero.
1629  const OptionalAmount &Amt = FS.getFieldWidth();
1630  if (Amt.getHowSpecified() == OptionalAmount::Constant) {
1631    if (Amt.getConstantAmount() == 0) {
1632      const CharSourceRange &R = getSpecifierRange(Amt.getStart(),
1633                                                   Amt.getConstantLength());
1634      S.Diag(getLocationOfByte(Amt.getStart()),
1635             diag::warn_scanf_nonzero_width)
1636        << R << FixItHint::CreateRemoval(R);
1637    }
1638  }
1639
1640  if (!FS.consumesDataArgument()) {
1641    // FIXME: Technically specifying a precision or field width here
1642    // makes no sense.  Worth issuing a warning at some point.
1643    return true;
1644  }
1645
1646  // Consume the argument.
1647  unsigned argIndex = FS.getArgIndex();
1648  if (argIndex < NumDataArgs) {
1649      // The check to see if the argIndex is valid will come later.
1650      // We set the bit here because we may exit early from this
1651      // function if we encounter some other error.
1652    CoveredArgs.set(argIndex);
1653  }
1654
1655  // FIXME: Check that the length modifier is valid with the given
1656  // conversion specifier.
1657
1658  // The remaining checks depend on the data arguments.
1659  if (HasVAListArg)
1660    return true;
1661
1662  if (argIndex >= NumDataArgs) {
1663    if (FS.usesPositionalArg())  {
1664      S.Diag(getLocationOfByte(CS.getStart()),
1665             diag::warn_printf_positional_arg_exceeds_data_args)
1666      << (argIndex+1) << NumDataArgs
1667      << getSpecifierRange(startSpecifier, specifierLen);
1668    }
1669    else {
1670      S.Diag(getLocationOfByte(CS.getStart()),
1671             diag::warn_printf_insufficient_data_args)
1672      << getSpecifierRange(startSpecifier, specifierLen);
1673    }
1674
1675    // Don't do any more checking.
1676    return false;
1677  }
1678
1679  // FIXME: Check that the argument type matches the format specifier.
1680
1681  return true;
1682}
1683
1684void Sema::CheckFormatString(const StringLiteral *FExpr,
1685                             const Expr *OrigFormatExpr,
1686                             const CallExpr *TheCall, bool HasVAListArg,
1687                             unsigned format_idx, unsigned firstDataArg,
1688                             bool isPrintf) {
1689
1690  // CHECK: is the format string a wide literal?
1691  if (FExpr->isWide()) {
1692    Diag(FExpr->getLocStart(),
1693         diag::warn_format_string_is_wide_literal)
1694    << OrigFormatExpr->getSourceRange();
1695    return;
1696  }
1697
1698  // Str - The format string.  NOTE: this is NOT null-terminated!
1699  const char *Str = FExpr->getStrData();
1700
1701  // CHECK: empty format string?
1702  unsigned StrLen = FExpr->getByteLength();
1703
1704  if (StrLen == 0) {
1705    Diag(FExpr->getLocStart(), diag::warn_empty_format_string)
1706    << OrigFormatExpr->getSourceRange();
1707    return;
1708  }
1709
1710  if (isPrintf) {
1711    CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1712                         TheCall->getNumArgs() - firstDataArg,
1713                         isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1714                         HasVAListArg, TheCall, format_idx);
1715
1716    if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen))
1717      H.DoneProcessing();
1718  }
1719  else {
1720    CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1721                        TheCall->getNumArgs() - firstDataArg,
1722                        isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1723                        HasVAListArg, TheCall, format_idx);
1724
1725    if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen))
1726      H.DoneProcessing();
1727  }
1728}
1729
1730//===--- CHECK: Return Address of Stack Variable --------------------------===//
1731
1732static DeclRefExpr* EvalVal(Expr *E);
1733static DeclRefExpr* EvalAddr(Expr* E);
1734
1735/// CheckReturnStackAddr - Check if a return statement returns the address
1736///   of a stack variable.
1737void
1738Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
1739                           SourceLocation ReturnLoc) {
1740
1741  // Perform checking for returned stack addresses.
1742  if (lhsType->isPointerType() || lhsType->isBlockPointerType()) {
1743    if (DeclRefExpr *DR = EvalAddr(RetValExp))
1744      Diag(DR->getLocStart(), diag::warn_ret_stack_addr)
1745       << DR->getDecl()->getDeclName() << RetValExp->getSourceRange();
1746
1747    // Skip over implicit cast expressions when checking for block expressions.
1748    RetValExp = RetValExp->IgnoreParenCasts();
1749
1750    if (BlockExpr *C = dyn_cast<BlockExpr>(RetValExp))
1751      if (C->hasBlockDeclRefExprs())
1752        Diag(C->getLocStart(), diag::err_ret_local_block)
1753          << C->getSourceRange();
1754
1755    if (AddrLabelExpr *ALE = dyn_cast<AddrLabelExpr>(RetValExp))
1756      Diag(ALE->getLocStart(), diag::warn_ret_addr_label)
1757        << ALE->getSourceRange();
1758
1759  } else if (lhsType->isReferenceType()) {
1760    // Perform checking for stack values returned by reference.
1761    // Check for a reference to the stack
1762    if (DeclRefExpr *DR = EvalVal(RetValExp))
1763      Diag(DR->getLocStart(), diag::warn_ret_stack_ref)
1764        << DR->getDecl()->getDeclName() << RetValExp->getSourceRange();
1765  }
1766}
1767
1768/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
1769///  check if the expression in a return statement evaluates to an address
1770///  to a location on the stack.  The recursion is used to traverse the
1771///  AST of the return expression, with recursion backtracking when we
1772///  encounter a subexpression that (1) clearly does not lead to the address
1773///  of a stack variable or (2) is something we cannot determine leads to
1774///  the address of a stack variable based on such local checking.
1775///
1776///  EvalAddr processes expressions that are pointers that are used as
1777///  references (and not L-values).  EvalVal handles all other values.
1778///  At the base case of the recursion is a check for a DeclRefExpr* in
1779///  the refers to a stack variable.
1780///
1781///  This implementation handles:
1782///
1783///   * pointer-to-pointer casts
1784///   * implicit conversions from array references to pointers
1785///   * taking the address of fields
1786///   * arbitrary interplay between "&" and "*" operators
1787///   * pointer arithmetic from an address of a stack variable
1788///   * taking the address of an array element where the array is on the stack
1789static DeclRefExpr* EvalAddr(Expr *E) {
1790  // We should only be called for evaluating pointer expressions.
1791  assert((E->getType()->isAnyPointerType() ||
1792          E->getType()->isBlockPointerType() ||
1793          E->getType()->isObjCQualifiedIdType()) &&
1794         "EvalAddr only works on pointers");
1795
1796  // Our "symbolic interpreter" is just a dispatch off the currently
1797  // viewed AST node.  We then recursively traverse the AST by calling
1798  // EvalAddr and EvalVal appropriately.
1799  switch (E->getStmtClass()) {
1800  case Stmt::ParenExprClass:
1801    // Ignore parentheses.
1802    return EvalAddr(cast<ParenExpr>(E)->getSubExpr());
1803
1804  case Stmt::UnaryOperatorClass: {
1805    // The only unary operator that make sense to handle here
1806    // is AddrOf.  All others don't make sense as pointers.
1807    UnaryOperator *U = cast<UnaryOperator>(E);
1808
1809    if (U->getOpcode() == UnaryOperator::AddrOf)
1810      return EvalVal(U->getSubExpr());
1811    else
1812      return NULL;
1813  }
1814
1815  case Stmt::BinaryOperatorClass: {
1816    // Handle pointer arithmetic.  All other binary operators are not valid
1817    // in this context.
1818    BinaryOperator *B = cast<BinaryOperator>(E);
1819    BinaryOperator::Opcode op = B->getOpcode();
1820
1821    if (op != BinaryOperator::Add && op != BinaryOperator::Sub)
1822      return NULL;
1823
1824    Expr *Base = B->getLHS();
1825
1826    // Determine which argument is the real pointer base.  It could be
1827    // the RHS argument instead of the LHS.
1828    if (!Base->getType()->isPointerType()) Base = B->getRHS();
1829
1830    assert (Base->getType()->isPointerType());
1831    return EvalAddr(Base);
1832  }
1833
1834  // For conditional operators we need to see if either the LHS or RHS are
1835  // valid DeclRefExpr*s.  If one of them is valid, we return it.
1836  case Stmt::ConditionalOperatorClass: {
1837    ConditionalOperator *C = cast<ConditionalOperator>(E);
1838
1839    // Handle the GNU extension for missing LHS.
1840    if (Expr *lhsExpr = C->getLHS())
1841      if (DeclRefExpr* LHS = EvalAddr(lhsExpr))
1842        return LHS;
1843
1844     return EvalAddr(C->getRHS());
1845  }
1846
1847  // For casts, we need to handle conversions from arrays to
1848  // pointer values, and pointer-to-pointer conversions.
1849  case Stmt::ImplicitCastExprClass:
1850  case Stmt::CStyleCastExprClass:
1851  case Stmt::CXXFunctionalCastExprClass: {
1852    Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
1853    QualType T = SubExpr->getType();
1854
1855    if (SubExpr->getType()->isPointerType() ||
1856        SubExpr->getType()->isBlockPointerType() ||
1857        SubExpr->getType()->isObjCQualifiedIdType())
1858      return EvalAddr(SubExpr);
1859    else if (T->isArrayType())
1860      return EvalVal(SubExpr);
1861    else
1862      return 0;
1863  }
1864
1865  // C++ casts.  For dynamic casts, static casts, and const casts, we
1866  // are always converting from a pointer-to-pointer, so we just blow
1867  // through the cast.  In the case the dynamic cast doesn't fail (and
1868  // return NULL), we take the conservative route and report cases
1869  // where we return the address of a stack variable.  For Reinterpre
1870  // FIXME: The comment about is wrong; we're not always converting
1871  // from pointer to pointer. I'm guessing that this code should also
1872  // handle references to objects.
1873  case Stmt::CXXStaticCastExprClass:
1874  case Stmt::CXXDynamicCastExprClass:
1875  case Stmt::CXXConstCastExprClass:
1876  case Stmt::CXXReinterpretCastExprClass: {
1877      Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr();
1878      if (S->getType()->isPointerType() || S->getType()->isBlockPointerType())
1879        return EvalAddr(S);
1880      else
1881        return NULL;
1882  }
1883
1884  // Everything else: we simply don't reason about them.
1885  default:
1886    return NULL;
1887  }
1888}
1889
1890
1891///  EvalVal - This function is complements EvalAddr in the mutual recursion.
1892///   See the comments for EvalAddr for more details.
1893static DeclRefExpr* EvalVal(Expr *E) {
1894
1895  // We should only be called for evaluating non-pointer expressions, or
1896  // expressions with a pointer type that are not used as references but instead
1897  // are l-values (e.g., DeclRefExpr with a pointer type).
1898
1899  // Our "symbolic interpreter" is just a dispatch off the currently
1900  // viewed AST node.  We then recursively traverse the AST by calling
1901  // EvalAddr and EvalVal appropriately.
1902  switch (E->getStmtClass()) {
1903  case Stmt::DeclRefExprClass: {
1904    // DeclRefExpr: the base case.  When we hit a DeclRefExpr we are looking
1905    //  at code that refers to a variable's name.  We check if it has local
1906    //  storage within the function, and if so, return the expression.
1907    DeclRefExpr *DR = cast<DeclRefExpr>(E);
1908
1909    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
1910      if (V->hasLocalStorage() && !V->getType()->isReferenceType()) return DR;
1911
1912    return NULL;
1913  }
1914
1915  case Stmt::ParenExprClass:
1916    // Ignore parentheses.
1917    return EvalVal(cast<ParenExpr>(E)->getSubExpr());
1918
1919  case Stmt::UnaryOperatorClass: {
1920    // The only unary operator that make sense to handle here
1921    // is Deref.  All others don't resolve to a "name."  This includes
1922    // handling all sorts of rvalues passed to a unary operator.
1923    UnaryOperator *U = cast<UnaryOperator>(E);
1924
1925    if (U->getOpcode() == UnaryOperator::Deref)
1926      return EvalAddr(U->getSubExpr());
1927
1928    return NULL;
1929  }
1930
1931  case Stmt::ArraySubscriptExprClass: {
1932    // Array subscripts are potential references to data on the stack.  We
1933    // retrieve the DeclRefExpr* for the array variable if it indeed
1934    // has local storage.
1935    return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase());
1936  }
1937
1938  case Stmt::ConditionalOperatorClass: {
1939    // For conditional operators we need to see if either the LHS or RHS are
1940    // non-NULL DeclRefExpr's.  If one is non-NULL, we return it.
1941    ConditionalOperator *C = cast<ConditionalOperator>(E);
1942
1943    // Handle the GNU extension for missing LHS.
1944    if (Expr *lhsExpr = C->getLHS())
1945      if (DeclRefExpr *LHS = EvalVal(lhsExpr))
1946        return LHS;
1947
1948    return EvalVal(C->getRHS());
1949  }
1950
1951  // Accesses to members are potential references to data on the stack.
1952  case Stmt::MemberExprClass: {
1953    MemberExpr *M = cast<MemberExpr>(E);
1954
1955    // Check for indirect access.  We only want direct field accesses.
1956    if (!M->isArrow())
1957      return EvalVal(M->getBase());
1958    else
1959      return NULL;
1960  }
1961
1962  // Everything else: we simply don't reason about them.
1963  default:
1964    return NULL;
1965  }
1966}
1967
1968//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
1969
1970/// Check for comparisons of floating point operands using != and ==.
1971/// Issue a warning if these are no self-comparisons, as they are not likely
1972/// to do what the programmer intended.
1973void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) {
1974  bool EmitWarning = true;
1975
1976  Expr* LeftExprSansParen = lex->IgnoreParens();
1977  Expr* RightExprSansParen = rex->IgnoreParens();
1978
1979  // Special case: check for x == x (which is OK).
1980  // Do not emit warnings for such cases.
1981  if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
1982    if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
1983      if (DRL->getDecl() == DRR->getDecl())
1984        EmitWarning = false;
1985
1986
1987  // Special case: check for comparisons against literals that can be exactly
1988  //  represented by APFloat.  In such cases, do not emit a warning.  This
1989  //  is a heuristic: often comparison against such literals are used to
1990  //  detect if a value in a variable has not changed.  This clearly can
1991  //  lead to false negatives.
1992  if (EmitWarning) {
1993    if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
1994      if (FLL->isExact())
1995        EmitWarning = false;
1996    } else
1997      if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){
1998        if (FLR->isExact())
1999          EmitWarning = false;
2000    }
2001  }
2002
2003  // Check for comparisons with builtin types.
2004  if (EmitWarning)
2005    if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
2006      if (CL->isBuiltinCall(Context))
2007        EmitWarning = false;
2008
2009  if (EmitWarning)
2010    if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
2011      if (CR->isBuiltinCall(Context))
2012        EmitWarning = false;
2013
2014  // Emit the diagnostic.
2015  if (EmitWarning)
2016    Diag(loc, diag::warn_floatingpoint_eq)
2017      << lex->getSourceRange() << rex->getSourceRange();
2018}
2019
2020//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===//
2021//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===//
2022
2023namespace {
2024
2025/// Structure recording the 'active' range of an integer-valued
2026/// expression.
2027struct IntRange {
2028  /// The number of bits active in the int.
2029  unsigned Width;
2030
2031  /// True if the int is known not to have negative values.
2032  bool NonNegative;
2033
2034  IntRange() {}
2035  IntRange(unsigned Width, bool NonNegative)
2036    : Width(Width), NonNegative(NonNegative)
2037  {}
2038
2039  // Returns the range of the bool type.
2040  static IntRange forBoolType() {
2041    return IntRange(1, true);
2042  }
2043
2044  // Returns the range of an integral type.
2045  static IntRange forType(ASTContext &C, QualType T) {
2046    return forCanonicalType(C, T->getCanonicalTypeInternal().getTypePtr());
2047  }
2048
2049  // Returns the range of an integeral type based on its canonical
2050  // representation.
2051  static IntRange forCanonicalType(ASTContext &C, const Type *T) {
2052    assert(T->isCanonicalUnqualified());
2053
2054    if (const VectorType *VT = dyn_cast<VectorType>(T))
2055      T = VT->getElementType().getTypePtr();
2056    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2057      T = CT->getElementType().getTypePtr();
2058
2059    if (const EnumType *ET = dyn_cast<EnumType>(T)) {
2060      EnumDecl *Enum = ET->getDecl();
2061      unsigned NumPositive = Enum->getNumPositiveBits();
2062      unsigned NumNegative = Enum->getNumNegativeBits();
2063
2064      return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0);
2065    }
2066
2067    const BuiltinType *BT = cast<BuiltinType>(T);
2068    assert(BT->isInteger());
2069
2070    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2071  }
2072
2073  // Returns the supremum of two ranges: i.e. their conservative merge.
2074  static IntRange join(IntRange L, IntRange R) {
2075    return IntRange(std::max(L.Width, R.Width),
2076                    L.NonNegative && R.NonNegative);
2077  }
2078
2079  // Returns the infinum of two ranges: i.e. their aggressive merge.
2080  static IntRange meet(IntRange L, IntRange R) {
2081    return IntRange(std::min(L.Width, R.Width),
2082                    L.NonNegative || R.NonNegative);
2083  }
2084};
2085
2086IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) {
2087  if (value.isSigned() && value.isNegative())
2088    return IntRange(value.getMinSignedBits(), false);
2089
2090  if (value.getBitWidth() > MaxWidth)
2091    value.trunc(MaxWidth);
2092
2093  // isNonNegative() just checks the sign bit without considering
2094  // signedness.
2095  return IntRange(value.getActiveBits(), true);
2096}
2097
2098IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty,
2099                       unsigned MaxWidth) {
2100  if (result.isInt())
2101    return GetValueRange(C, result.getInt(), MaxWidth);
2102
2103  if (result.isVector()) {
2104    IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth);
2105    for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) {
2106      IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth);
2107      R = IntRange::join(R, El);
2108    }
2109    return R;
2110  }
2111
2112  if (result.isComplexInt()) {
2113    IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth);
2114    IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth);
2115    return IntRange::join(R, I);
2116  }
2117
2118  // This can happen with lossless casts to intptr_t of "based" lvalues.
2119  // Assume it might use arbitrary bits.
2120  // FIXME: The only reason we need to pass the type in here is to get
2121  // the sign right on this one case.  It would be nice if APValue
2122  // preserved this.
2123  assert(result.isLValue());
2124  return IntRange(MaxWidth, Ty->isUnsignedIntegerType());
2125}
2126
2127/// Pseudo-evaluate the given integer expression, estimating the
2128/// range of values it might take.
2129///
2130/// \param MaxWidth - the width to which the value will be truncated
2131IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) {
2132  E = E->IgnoreParens();
2133
2134  // Try a full evaluation first.
2135  Expr::EvalResult result;
2136  if (E->Evaluate(result, C))
2137    return GetValueRange(C, result.Val, E->getType(), MaxWidth);
2138
2139  // I think we only want to look through implicit casts here; if the
2140  // user has an explicit widening cast, we should treat the value as
2141  // being of the new, wider type.
2142  if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) {
2143    if (CE->getCastKind() == CastExpr::CK_NoOp)
2144      return GetExprRange(C, CE->getSubExpr(), MaxWidth);
2145
2146    IntRange OutputTypeRange = IntRange::forType(C, CE->getType());
2147
2148    bool isIntegerCast = (CE->getCastKind() == CastExpr::CK_IntegralCast);
2149    if (!isIntegerCast && CE->getCastKind() == CastExpr::CK_Unknown)
2150      isIntegerCast = CE->getSubExpr()->getType()->isIntegerType();
2151
2152    // Assume that non-integer casts can span the full range of the type.
2153    if (!isIntegerCast)
2154      return OutputTypeRange;
2155
2156    IntRange SubRange
2157      = GetExprRange(C, CE->getSubExpr(),
2158                     std::min(MaxWidth, OutputTypeRange.Width));
2159
2160    // Bail out if the subexpr's range is as wide as the cast type.
2161    if (SubRange.Width >= OutputTypeRange.Width)
2162      return OutputTypeRange;
2163
2164    // Otherwise, we take the smaller width, and we're non-negative if
2165    // either the output type or the subexpr is.
2166    return IntRange(SubRange.Width,
2167                    SubRange.NonNegative || OutputTypeRange.NonNegative);
2168  }
2169
2170  if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
2171    // If we can fold the condition, just take that operand.
2172    bool CondResult;
2173    if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C))
2174      return GetExprRange(C, CondResult ? CO->getTrueExpr()
2175                                        : CO->getFalseExpr(),
2176                          MaxWidth);
2177
2178    // Otherwise, conservatively merge.
2179    IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth);
2180    IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth);
2181    return IntRange::join(L, R);
2182  }
2183
2184  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
2185    switch (BO->getOpcode()) {
2186
2187    // Boolean-valued operations are single-bit and positive.
2188    case BinaryOperator::LAnd:
2189    case BinaryOperator::LOr:
2190    case BinaryOperator::LT:
2191    case BinaryOperator::GT:
2192    case BinaryOperator::LE:
2193    case BinaryOperator::GE:
2194    case BinaryOperator::EQ:
2195    case BinaryOperator::NE:
2196      return IntRange::forBoolType();
2197
2198    // The type of these compound assignments is the type of the LHS,
2199    // so the RHS is not necessarily an integer.
2200    case BinaryOperator::MulAssign:
2201    case BinaryOperator::DivAssign:
2202    case BinaryOperator::RemAssign:
2203    case BinaryOperator::AddAssign:
2204    case BinaryOperator::SubAssign:
2205      return IntRange::forType(C, E->getType());
2206
2207    // Operations with opaque sources are black-listed.
2208    case BinaryOperator::PtrMemD:
2209    case BinaryOperator::PtrMemI:
2210      return IntRange::forType(C, E->getType());
2211
2212    // Bitwise-and uses the *infinum* of the two source ranges.
2213    case BinaryOperator::And:
2214    case BinaryOperator::AndAssign:
2215      return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth),
2216                            GetExprRange(C, BO->getRHS(), MaxWidth));
2217
2218    // Left shift gets black-listed based on a judgement call.
2219    case BinaryOperator::Shl:
2220      // ...except that we want to treat '1 << (blah)' as logically
2221      // positive.  It's an important idiom.
2222      if (IntegerLiteral *I
2223            = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) {
2224        if (I->getValue() == 1) {
2225          IntRange R = IntRange::forType(C, E->getType());
2226          return IntRange(R.Width, /*NonNegative*/ true);
2227        }
2228      }
2229      // fallthrough
2230
2231    case BinaryOperator::ShlAssign:
2232      return IntRange::forType(C, E->getType());
2233
2234    // Right shift by a constant can narrow its left argument.
2235    case BinaryOperator::Shr:
2236    case BinaryOperator::ShrAssign: {
2237      IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2238
2239      // If the shift amount is a positive constant, drop the width by
2240      // that much.
2241      llvm::APSInt shift;
2242      if (BO->getRHS()->isIntegerConstantExpr(shift, C) &&
2243          shift.isNonNegative()) {
2244        unsigned zext = shift.getZExtValue();
2245        if (zext >= L.Width)
2246          L.Width = (L.NonNegative ? 0 : 1);
2247        else
2248          L.Width -= zext;
2249      }
2250
2251      return L;
2252    }
2253
2254    // Comma acts as its right operand.
2255    case BinaryOperator::Comma:
2256      return GetExprRange(C, BO->getRHS(), MaxWidth);
2257
2258    // Black-list pointer subtractions.
2259    case BinaryOperator::Sub:
2260      if (BO->getLHS()->getType()->isPointerType())
2261        return IntRange::forType(C, E->getType());
2262      // fallthrough
2263
2264    default:
2265      break;
2266    }
2267
2268    // Treat every other operator as if it were closed on the
2269    // narrowest type that encompasses both operands.
2270    IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2271    IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth);
2272    return IntRange::join(L, R);
2273  }
2274
2275  if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
2276    switch (UO->getOpcode()) {
2277    // Boolean-valued operations are white-listed.
2278    case UnaryOperator::LNot:
2279      return IntRange::forBoolType();
2280
2281    // Operations with opaque sources are black-listed.
2282    case UnaryOperator::Deref:
2283    case UnaryOperator::AddrOf: // should be impossible
2284    case UnaryOperator::OffsetOf:
2285      return IntRange::forType(C, E->getType());
2286
2287    default:
2288      return GetExprRange(C, UO->getSubExpr(), MaxWidth);
2289    }
2290  }
2291
2292  if (dyn_cast<OffsetOfExpr>(E)) {
2293    IntRange::forType(C, E->getType());
2294  }
2295
2296  FieldDecl *BitField = E->getBitField();
2297  if (BitField) {
2298    llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C);
2299    unsigned BitWidth = BitWidthAP.getZExtValue();
2300
2301    return IntRange(BitWidth, BitField->getType()->isUnsignedIntegerType());
2302  }
2303
2304  return IntRange::forType(C, E->getType());
2305}
2306
2307IntRange GetExprRange(ASTContext &C, Expr *E) {
2308  return GetExprRange(C, E, C.getIntWidth(E->getType()));
2309}
2310
2311/// Checks whether the given value, which currently has the given
2312/// source semantics, has the same value when coerced through the
2313/// target semantics.
2314bool IsSameFloatAfterCast(const llvm::APFloat &value,
2315                          const llvm::fltSemantics &Src,
2316                          const llvm::fltSemantics &Tgt) {
2317  llvm::APFloat truncated = value;
2318
2319  bool ignored;
2320  truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored);
2321  truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored);
2322
2323  return truncated.bitwiseIsEqual(value);
2324}
2325
2326/// Checks whether the given value, which currently has the given
2327/// source semantics, has the same value when coerced through the
2328/// target semantics.
2329///
2330/// The value might be a vector of floats (or a complex number).
2331bool IsSameFloatAfterCast(const APValue &value,
2332                          const llvm::fltSemantics &Src,
2333                          const llvm::fltSemantics &Tgt) {
2334  if (value.isFloat())
2335    return IsSameFloatAfterCast(value.getFloat(), Src, Tgt);
2336
2337  if (value.isVector()) {
2338    for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i)
2339      if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt))
2340        return false;
2341    return true;
2342  }
2343
2344  assert(value.isComplexFloat());
2345  return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) &&
2346          IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt));
2347}
2348
2349void AnalyzeImplicitConversions(Sema &S, Expr *E);
2350
2351bool IsZero(Sema &S, Expr *E) {
2352  llvm::APSInt Value;
2353  return E->isIntegerConstantExpr(Value, S.Context) && Value == 0;
2354}
2355
2356void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) {
2357  BinaryOperator::Opcode op = E->getOpcode();
2358  if (op == BinaryOperator::LT && IsZero(S, E->getRHS())) {
2359    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2360      << "< 0" << "false"
2361      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2362  } else if (op == BinaryOperator::GE && IsZero(S, E->getRHS())) {
2363    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2364      << ">= 0" << "true"
2365      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2366  } else if (op == BinaryOperator::GT && IsZero(S, E->getLHS())) {
2367    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
2368      << "0 >" << "false"
2369      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2370  } else if (op == BinaryOperator::LE && IsZero(S, E->getLHS())) {
2371    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
2372      << "0 <=" << "true"
2373      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2374  }
2375}
2376
2377/// Analyze the operands of the given comparison.  Implements the
2378/// fallback case from AnalyzeComparison.
2379void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) {
2380  AnalyzeImplicitConversions(S, E->getLHS());
2381  AnalyzeImplicitConversions(S, E->getRHS());
2382}
2383
2384/// \brief Implements -Wsign-compare.
2385///
2386/// \param lex the left-hand expression
2387/// \param rex the right-hand expression
2388/// \param OpLoc the location of the joining operator
2389/// \param BinOpc binary opcode or 0
2390void AnalyzeComparison(Sema &S, BinaryOperator *E) {
2391  // The type the comparison is being performed in.
2392  QualType T = E->getLHS()->getType();
2393  assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType())
2394         && "comparison with mismatched types");
2395
2396  // We don't do anything special if this isn't an unsigned integral
2397  // comparison:  we're only interested in integral comparisons, and
2398  // signed comparisons only happen in cases we don't care to warn about.
2399  if (!T->isUnsignedIntegerType())
2400    return AnalyzeImpConvsInComparison(S, E);
2401
2402  Expr *lex = E->getLHS()->IgnoreParenImpCasts();
2403  Expr *rex = E->getRHS()->IgnoreParenImpCasts();
2404
2405  // Check to see if one of the (unmodified) operands is of different
2406  // signedness.
2407  Expr *signedOperand, *unsignedOperand;
2408  if (lex->getType()->isSignedIntegerType()) {
2409    assert(!rex->getType()->isSignedIntegerType() &&
2410           "unsigned comparison between two signed integer expressions?");
2411    signedOperand = lex;
2412    unsignedOperand = rex;
2413  } else if (rex->getType()->isSignedIntegerType()) {
2414    signedOperand = rex;
2415    unsignedOperand = lex;
2416  } else {
2417    CheckTrivialUnsignedComparison(S, E);
2418    return AnalyzeImpConvsInComparison(S, E);
2419  }
2420
2421  // Otherwise, calculate the effective range of the signed operand.
2422  IntRange signedRange = GetExprRange(S.Context, signedOperand);
2423
2424  // Go ahead and analyze implicit conversions in the operands.  Note
2425  // that we skip the implicit conversions on both sides.
2426  AnalyzeImplicitConversions(S, lex);
2427  AnalyzeImplicitConversions(S, rex);
2428
2429  // If the signed range is non-negative, -Wsign-compare won't fire,
2430  // but we should still check for comparisons which are always true
2431  // or false.
2432  if (signedRange.NonNegative)
2433    return CheckTrivialUnsignedComparison(S, E);
2434
2435  // For (in)equality comparisons, if the unsigned operand is a
2436  // constant which cannot collide with a overflowed signed operand,
2437  // then reinterpreting the signed operand as unsigned will not
2438  // change the result of the comparison.
2439  if (E->isEqualityOp()) {
2440    unsigned comparisonWidth = S.Context.getIntWidth(T);
2441    IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand);
2442
2443    // We should never be unable to prove that the unsigned operand is
2444    // non-negative.
2445    assert(unsignedRange.NonNegative && "unsigned range includes negative?");
2446
2447    if (unsignedRange.Width < comparisonWidth)
2448      return;
2449  }
2450
2451  S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison)
2452    << lex->getType() << rex->getType()
2453    << lex->getSourceRange() << rex->getSourceRange();
2454}
2455
2456/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
2457void DiagnoseImpCast(Sema &S, Expr *E, QualType T, unsigned diag) {
2458  S.Diag(E->getExprLoc(), diag) << E->getType() << T << E->getSourceRange();
2459}
2460
2461void CheckImplicitConversion(Sema &S, Expr *E, QualType T,
2462                             bool *ICContext = 0) {
2463  if (E->isTypeDependent() || E->isValueDependent()) return;
2464
2465  const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr();
2466  const Type *Target = S.Context.getCanonicalType(T).getTypePtr();
2467  if (Source == Target) return;
2468  if (Target->isDependentType()) return;
2469
2470  // Never diagnose implicit casts to bool.
2471  if (Target->isSpecificBuiltinType(BuiltinType::Bool))
2472    return;
2473
2474  // Strip vector types.
2475  if (isa<VectorType>(Source)) {
2476    if (!isa<VectorType>(Target))
2477      return DiagnoseImpCast(S, E, T, diag::warn_impcast_vector_scalar);
2478
2479    Source = cast<VectorType>(Source)->getElementType().getTypePtr();
2480    Target = cast<VectorType>(Target)->getElementType().getTypePtr();
2481  }
2482
2483  // Strip complex types.
2484  if (isa<ComplexType>(Source)) {
2485    if (!isa<ComplexType>(Target))
2486      return DiagnoseImpCast(S, E, T, diag::warn_impcast_complex_scalar);
2487
2488    Source = cast<ComplexType>(Source)->getElementType().getTypePtr();
2489    Target = cast<ComplexType>(Target)->getElementType().getTypePtr();
2490  }
2491
2492  const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source);
2493  const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target);
2494
2495  // If the source is floating point...
2496  if (SourceBT && SourceBT->isFloatingPoint()) {
2497    // ...and the target is floating point...
2498    if (TargetBT && TargetBT->isFloatingPoint()) {
2499      // ...then warn if we're dropping FP rank.
2500
2501      // Builtin FP kinds are ordered by increasing FP rank.
2502      if (SourceBT->getKind() > TargetBT->getKind()) {
2503        // Don't warn about float constants that are precisely
2504        // representable in the target type.
2505        Expr::EvalResult result;
2506        if (E->Evaluate(result, S.Context)) {
2507          // Value might be a float, a float vector, or a float complex.
2508          if (IsSameFloatAfterCast(result.Val,
2509                   S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)),
2510                   S.Context.getFloatTypeSemantics(QualType(SourceBT, 0))))
2511            return;
2512        }
2513
2514        DiagnoseImpCast(S, E, T, diag::warn_impcast_float_precision);
2515      }
2516      return;
2517    }
2518
2519    // If the target is integral, always warn.
2520    if ((TargetBT && TargetBT->isInteger()))
2521      // TODO: don't warn for integer values?
2522      DiagnoseImpCast(S, E, T, diag::warn_impcast_float_integer);
2523
2524    return;
2525  }
2526
2527  if (!Source->isIntegerType() || !Target->isIntegerType())
2528    return;
2529
2530  IntRange SourceRange = GetExprRange(S.Context, E);
2531  IntRange TargetRange = IntRange::forCanonicalType(S.Context, Target);
2532
2533  if (SourceRange.Width > TargetRange.Width) {
2534    // People want to build with -Wshorten-64-to-32 and not -Wconversion
2535    // and by god we'll let them.
2536    if (SourceRange.Width == 64 && TargetRange.Width == 32)
2537      return DiagnoseImpCast(S, E, T, diag::warn_impcast_integer_64_32);
2538    return DiagnoseImpCast(S, E, T, diag::warn_impcast_integer_precision);
2539  }
2540
2541  if ((TargetRange.NonNegative && !SourceRange.NonNegative) ||
2542      (!TargetRange.NonNegative && SourceRange.NonNegative &&
2543       SourceRange.Width == TargetRange.Width)) {
2544    unsigned DiagID = diag::warn_impcast_integer_sign;
2545
2546    // Traditionally, gcc has warned about this under -Wsign-compare.
2547    // We also want to warn about it in -Wconversion.
2548    // So if -Wconversion is off, use a completely identical diagnostic
2549    // in the sign-compare group.
2550    // The conditional-checking code will
2551    if (ICContext) {
2552      DiagID = diag::warn_impcast_integer_sign_conditional;
2553      *ICContext = true;
2554    }
2555
2556    return DiagnoseImpCast(S, E, T, DiagID);
2557  }
2558
2559  return;
2560}
2561
2562void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T);
2563
2564void CheckConditionalOperand(Sema &S, Expr *E, QualType T,
2565                             bool &ICContext) {
2566  E = E->IgnoreParenImpCasts();
2567
2568  if (isa<ConditionalOperator>(E))
2569    return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T);
2570
2571  AnalyzeImplicitConversions(S, E);
2572  if (E->getType() != T)
2573    return CheckImplicitConversion(S, E, T, &ICContext);
2574  return;
2575}
2576
2577void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) {
2578  AnalyzeImplicitConversions(S, E->getCond());
2579
2580  bool Suspicious = false;
2581  CheckConditionalOperand(S, E->getTrueExpr(), T, Suspicious);
2582  CheckConditionalOperand(S, E->getFalseExpr(), T, Suspicious);
2583
2584  // If -Wconversion would have warned about either of the candidates
2585  // for a signedness conversion to the context type...
2586  if (!Suspicious) return;
2587
2588  // ...but it's currently ignored...
2589  if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional))
2590    return;
2591
2592  // ...and -Wsign-compare isn't...
2593  if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional))
2594    return;
2595
2596  // ...then check whether it would have warned about either of the
2597  // candidates for a signedness conversion to the condition type.
2598  if (E->getType() != T) {
2599    Suspicious = false;
2600    CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(),
2601                            E->getType(), &Suspicious);
2602    if (!Suspicious)
2603      CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(),
2604                              E->getType(), &Suspicious);
2605    if (!Suspicious)
2606      return;
2607  }
2608
2609  // If so, emit a diagnostic under -Wsign-compare.
2610  Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts();
2611  Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts();
2612  S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional)
2613    << lex->getType() << rex->getType()
2614    << lex->getSourceRange() << rex->getSourceRange();
2615}
2616
2617/// AnalyzeImplicitConversions - Find and report any interesting
2618/// implicit conversions in the given expression.  There are a couple
2619/// of competing diagnostics here, -Wconversion and -Wsign-compare.
2620void AnalyzeImplicitConversions(Sema &S, Expr *OrigE) {
2621  QualType T = OrigE->getType();
2622  Expr *E = OrigE->IgnoreParenImpCasts();
2623
2624  // For conditional operators, we analyze the arguments as if they
2625  // were being fed directly into the output.
2626  if (isa<ConditionalOperator>(E)) {
2627    ConditionalOperator *CO = cast<ConditionalOperator>(E);
2628    CheckConditionalOperator(S, CO, T);
2629    return;
2630  }
2631
2632  // Go ahead and check any implicit conversions we might have skipped.
2633  // The non-canonical typecheck is just an optimization;
2634  // CheckImplicitConversion will filter out dead implicit conversions.
2635  if (E->getType() != T)
2636    CheckImplicitConversion(S, E, T);
2637
2638  // Now continue drilling into this expression.
2639
2640  // Skip past explicit casts.
2641  if (isa<ExplicitCastExpr>(E)) {
2642    E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts();
2643    return AnalyzeImplicitConversions(S, E);
2644  }
2645
2646  // Do a somewhat different check with comparison operators.
2647  if (isa<BinaryOperator>(E) && cast<BinaryOperator>(E)->isComparisonOp())
2648    return AnalyzeComparison(S, cast<BinaryOperator>(E));
2649
2650  // These break the otherwise-useful invariant below.  Fortunately,
2651  // we don't really need to recurse into them, because any internal
2652  // expressions should have been analyzed already when they were
2653  // built into statements.
2654  if (isa<StmtExpr>(E)) return;
2655
2656  // Don't descend into unevaluated contexts.
2657  if (isa<SizeOfAlignOfExpr>(E)) return;
2658
2659  // Now just recurse over the expression's children.
2660  for (Stmt::child_iterator I = E->child_begin(), IE = E->child_end();
2661         I != IE; ++I)
2662    AnalyzeImplicitConversions(S, cast<Expr>(*I));
2663}
2664
2665} // end anonymous namespace
2666
2667/// Diagnoses "dangerous" implicit conversions within the given
2668/// expression (which is a full expression).  Implements -Wconversion
2669/// and -Wsign-compare.
2670void Sema::CheckImplicitConversions(Expr *E) {
2671  // Don't diagnose in unevaluated contexts.
2672  if (ExprEvalContexts.back().Context == Sema::Unevaluated)
2673    return;
2674
2675  // Don't diagnose for value- or type-dependent expressions.
2676  if (E->isTypeDependent() || E->isValueDependent())
2677    return;
2678
2679  AnalyzeImplicitConversions(*this, E);
2680}
2681
2682/// CheckParmsForFunctionDef - Check that the parameters of the given
2683/// function are appropriate for the definition of a function. This
2684/// takes care of any checks that cannot be performed on the
2685/// declaration itself, e.g., that the types of each of the function
2686/// parameters are complete.
2687bool Sema::CheckParmsForFunctionDef(FunctionDecl *FD) {
2688  bool HasInvalidParm = false;
2689  for (unsigned p = 0, NumParams = FD->getNumParams(); p < NumParams; ++p) {
2690    ParmVarDecl *Param = FD->getParamDecl(p);
2691
2692    // C99 6.7.5.3p4: the parameters in a parameter type list in a
2693    // function declarator that is part of a function definition of
2694    // that function shall not have incomplete type.
2695    //
2696    // This is also C++ [dcl.fct]p6.
2697    if (!Param->isInvalidDecl() &&
2698        RequireCompleteType(Param->getLocation(), Param->getType(),
2699                               diag::err_typecheck_decl_incomplete_type)) {
2700      Param->setInvalidDecl();
2701      HasInvalidParm = true;
2702    }
2703
2704    // C99 6.9.1p5: If the declarator includes a parameter type list, the
2705    // declaration of each parameter shall include an identifier.
2706    if (Param->getIdentifier() == 0 &&
2707        !Param->isImplicit() &&
2708        !getLangOptions().CPlusPlus)
2709      Diag(Param->getLocation(), diag::err_parameter_name_omitted);
2710
2711    // C99 6.7.5.3p12:
2712    //   If the function declarator is not part of a definition of that
2713    //   function, parameters may have incomplete type and may use the [*]
2714    //   notation in their sequences of declarator specifiers to specify
2715    //   variable length array types.
2716    QualType PType = Param->getOriginalType();
2717    if (const ArrayType *AT = Context.getAsArrayType(PType)) {
2718      if (AT->getSizeModifier() == ArrayType::Star) {
2719        // FIXME: This diagnosic should point the the '[*]' if source-location
2720        // information is added for it.
2721        Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
2722      }
2723    }
2724  }
2725
2726  return HasInvalidParm;
2727}
2728