SemaChecking.cpp revision 000d428347f352979e0f6dffcf0a64e73af0a2b5
1eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch//===--- SemaChecking.cpp - Extra Semantic Checking -----------------------===//
2eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch//
3eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch//                     The LLVM Compiler Infrastructure
4eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch//
5eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch// This file is distributed under the University of Illinois Open Source
6eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch// License. See LICENSE.TXT for details.
7eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch//
8eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch//===----------------------------------------------------------------------===//
9eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch//
10eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch//  This file implements extra semantic analysis beyond what is enforced
11eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch//  by the C type system.
12eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch//
13eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch//===----------------------------------------------------------------------===//
14eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch
15eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "clang/Sema/Sema.h"
16eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "clang/Sema/SemaInternal.h"
17eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "clang/Sema/ScopeInfo.h"
18eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "clang/Analysis/Analyses/FormatString.h"
19eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "clang/AST/ASTContext.h"
20eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "clang/AST/CharUnits.h"
21eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "clang/AST/DeclCXX.h"
22eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "clang/AST/DeclObjC.h"
23eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "clang/AST/ExprCXX.h"
24eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "clang/AST/ExprObjC.h"
25eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "clang/AST/EvaluatedExprVisitor.h"
26eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "clang/AST/DeclObjC.h"
27eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "clang/AST/StmtCXX.h"
28eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "clang/AST/StmtObjC.h"
29eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "clang/Lex/Preprocessor.h"
30eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "llvm/ADT/BitVector.h"
31eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "llvm/ADT/STLExtras.h"
32eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "llvm/Support/raw_ostream.h"
33eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "clang/Basic/TargetBuiltins.h"
34eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "clang/Basic/TargetInfo.h"
35eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include "clang/Basic/ConvertUTF.h"
36eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch#include <limits>
37eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdochusing namespace clang;
387dbb3d5cf0c15f500944d211057644d6a2f37371Ben Murdochusing namespace sema;
397dbb3d5cf0c15f500944d211057644d6a2f37371Ben Murdoch
407dbb3d5cf0c15f500944d211057644d6a2f37371Ben MurdochSourceLocation Sema::getLocationOfStringLiteralByte(const StringLiteral *SL,
417dbb3d5cf0c15f500944d211057644d6a2f37371Ben Murdoch                                                    unsigned ByteNo) const {
42eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch  return SL->getLocationOfByte(ByteNo, PP.getSourceManager(),
43eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch                               PP.getLangOptions(), PP.getTargetInfo());
44eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch}
457dbb3d5cf0c15f500944d211057644d6a2f37371Ben Murdoch
467dbb3d5cf0c15f500944d211057644d6a2f37371Ben Murdoch
477dbb3d5cf0c15f500944d211057644d6a2f37371Ben Murdoch/// CheckablePrintfAttr - does a function call have a "printf" attribute
487dbb3d5cf0c15f500944d211057644d6a2f37371Ben Murdoch/// and arguments that merit checking?
497dbb3d5cf0c15f500944d211057644d6a2f37371Ben Murdochbool Sema::CheckablePrintfAttr(const FormatAttr *Format, CallExpr *TheCall) {
507dbb3d5cf0c15f500944d211057644d6a2f37371Ben Murdoch  if (Format->getType() == "printf") return true;
517dbb3d5cf0c15f500944d211057644d6a2f37371Ben Murdoch  if (Format->getType() == "printf0") {
52eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch    // printf0 allows null "format" string; if so don't check format/args
53eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch    unsigned format_idx = Format->getFormatIdx() - 1;
54eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch    // Does the index refer to the implicit object argument?
55eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch    if (isa<CXXMemberCallExpr>(TheCall)) {
56eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch      if (format_idx == 0)
57eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch        return false;
58eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch      --format_idx;
59eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch    }
60eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch    if (format_idx < TheCall->getNumArgs()) {
61eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch      Expr *Format = TheCall->getArg(format_idx)->IgnoreParenCasts();
62eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch      if (!Format->isNullPointerConstant(Context,
63eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch                                         Expr::NPC_ValueDependentIsNull))
64eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch        return true;
65eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch    }
66eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch  }
67eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch  return false;
68eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch}
69eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch
70eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch/// Checks that a call expression's argument count is the desired number.
71eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch/// This is useful when doing custom type-checking.  Returns true on error.
72eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdochstatic bool checkArgCount(Sema &S, CallExpr *call, unsigned desiredArgCount) {
73eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch  unsigned argCount = call->getNumArgs();
74eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch  if (argCount == desiredArgCount) return false;
75eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch
767dbb3d5cf0c15f500944d211057644d6a2f37371Ben Murdoch  if (argCount < desiredArgCount)
777dbb3d5cf0c15f500944d211057644d6a2f37371Ben Murdoch    return S.Diag(call->getLocEnd(), diag::err_typecheck_call_too_few_args)
78eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch        << 0 /*function call*/ << desiredArgCount << argCount
79eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch        << call->getSourceRange();
807dbb3d5cf0c15f500944d211057644d6a2f37371Ben Murdoch
817dbb3d5cf0c15f500944d211057644d6a2f37371Ben Murdoch  // Highlight all the excess arguments.
82eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch  SourceRange range(call->getArg(desiredArgCount)->getLocStart(),
83eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch                    call->getArg(argCount - 1)->getLocEnd());
84eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch
85eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch  return S.Diag(range.getBegin(), diag::err_typecheck_call_too_many_args)
86eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch    << 0 /*function call*/ << desiredArgCount << argCount
87eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch    << call->getArg(1)->getSourceRange();
88eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch}
89eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch
90eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen MurdochExprResult
91eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen MurdochSema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
92eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch  ExprResult TheCallResult(Owned(TheCall));
93eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch
94eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch  // Find out if any arguments are required to be integer constant expressions.
95eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch  unsigned ICEArguments = 0;
96eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch  ASTContext::GetBuiltinTypeError Error;
97eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch  Context.GetBuiltinType(BuiltinID, Error, &ICEArguments);
98eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch  if (Error != ASTContext::GE_None)
99eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch    ICEArguments = 0;  // Don't diagnose previously diagnosed errors.
100eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch
101eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch  // If any arguments are required to be ICE's, check and diagnose.
102eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch  for (unsigned ArgNo = 0; ICEArguments != 0; ++ArgNo) {
103eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch    // Skip arguments not required to be ICE's.
104eb525c5499e34cc9c4b825d6d9e75bb07cc06aceBen Murdoch    if ((ICEArguments & (1 << ArgNo)) == 0) continue;
105
106    llvm::APSInt Result;
107    if (SemaBuiltinConstantArg(TheCall, ArgNo, Result))
108      return true;
109    ICEArguments &= ~(1 << ArgNo);
110  }
111
112  switch (BuiltinID) {
113  case Builtin::BI__builtin___CFStringMakeConstantString:
114    assert(TheCall->getNumArgs() == 1 &&
115           "Wrong # arguments to builtin CFStringMakeConstantString");
116    if (CheckObjCString(TheCall->getArg(0)))
117      return ExprError();
118    break;
119  case Builtin::BI__builtin_stdarg_start:
120  case Builtin::BI__builtin_va_start:
121    if (SemaBuiltinVAStart(TheCall))
122      return ExprError();
123    break;
124  case Builtin::BI__builtin_isgreater:
125  case Builtin::BI__builtin_isgreaterequal:
126  case Builtin::BI__builtin_isless:
127  case Builtin::BI__builtin_islessequal:
128  case Builtin::BI__builtin_islessgreater:
129  case Builtin::BI__builtin_isunordered:
130    if (SemaBuiltinUnorderedCompare(TheCall))
131      return ExprError();
132    break;
133  case Builtin::BI__builtin_fpclassify:
134    if (SemaBuiltinFPClassification(TheCall, 6))
135      return ExprError();
136    break;
137  case Builtin::BI__builtin_isfinite:
138  case Builtin::BI__builtin_isinf:
139  case Builtin::BI__builtin_isinf_sign:
140  case Builtin::BI__builtin_isnan:
141  case Builtin::BI__builtin_isnormal:
142    if (SemaBuiltinFPClassification(TheCall, 1))
143      return ExprError();
144    break;
145  case Builtin::BI__builtin_shufflevector:
146    return SemaBuiltinShuffleVector(TheCall);
147    // TheCall will be freed by the smart pointer here, but that's fine, since
148    // SemaBuiltinShuffleVector guts it, but then doesn't release it.
149  case Builtin::BI__builtin_prefetch:
150    if (SemaBuiltinPrefetch(TheCall))
151      return ExprError();
152    break;
153  case Builtin::BI__builtin_object_size:
154    if (SemaBuiltinObjectSize(TheCall))
155      return ExprError();
156    break;
157  case Builtin::BI__builtin_longjmp:
158    if (SemaBuiltinLongjmp(TheCall))
159      return ExprError();
160    break;
161
162  case Builtin::BI__builtin_classify_type:
163    if (checkArgCount(*this, TheCall, 1)) return true;
164    TheCall->setType(Context.IntTy);
165    break;
166  case Builtin::BI__builtin_constant_p:
167    if (checkArgCount(*this, TheCall, 1)) return true;
168    TheCall->setType(Context.IntTy);
169    break;
170  case Builtin::BI__sync_fetch_and_add:
171  case Builtin::BI__sync_fetch_and_sub:
172  case Builtin::BI__sync_fetch_and_or:
173  case Builtin::BI__sync_fetch_and_and:
174  case Builtin::BI__sync_fetch_and_xor:
175  case Builtin::BI__sync_add_and_fetch:
176  case Builtin::BI__sync_sub_and_fetch:
177  case Builtin::BI__sync_and_and_fetch:
178  case Builtin::BI__sync_or_and_fetch:
179  case Builtin::BI__sync_xor_and_fetch:
180  case Builtin::BI__sync_val_compare_and_swap:
181  case Builtin::BI__sync_bool_compare_and_swap:
182  case Builtin::BI__sync_lock_test_and_set:
183  case Builtin::BI__sync_lock_release:
184  case Builtin::BI__sync_swap:
185    return SemaBuiltinAtomicOverloaded(move(TheCallResult));
186  }
187
188  // Since the target specific builtins for each arch overlap, only check those
189  // of the arch we are compiling for.
190  if (BuiltinID >= Builtin::FirstTSBuiltin) {
191    switch (Context.Target.getTriple().getArch()) {
192      case llvm::Triple::arm:
193      case llvm::Triple::thumb:
194        if (CheckARMBuiltinFunctionCall(BuiltinID, TheCall))
195          return ExprError();
196        break;
197      default:
198        break;
199    }
200  }
201
202  return move(TheCallResult);
203}
204
205// Get the valid immediate range for the specified NEON type code.
206static unsigned RFT(unsigned t, bool shift = false) {
207  bool quad = t & 0x10;
208
209  switch (t & 0x7) {
210    case 0: // i8
211      return shift ? 7 : (8 << (int)quad) - 1;
212    case 1: // i16
213      return shift ? 15 : (4 << (int)quad) - 1;
214    case 2: // i32
215      return shift ? 31 : (2 << (int)quad) - 1;
216    case 3: // i64
217      return shift ? 63 : (1 << (int)quad) - 1;
218    case 4: // f32
219      assert(!shift && "cannot shift float types!");
220      return (2 << (int)quad) - 1;
221    case 5: // poly8
222      return shift ? 7 : (8 << (int)quad) - 1;
223    case 6: // poly16
224      return shift ? 15 : (4 << (int)quad) - 1;
225    case 7: // float16
226      assert(!shift && "cannot shift float types!");
227      return (4 << (int)quad) - 1;
228  }
229  return 0;
230}
231
232bool Sema::CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
233  llvm::APSInt Result;
234
235  unsigned mask = 0;
236  unsigned TV = 0;
237  switch (BuiltinID) {
238#define GET_NEON_OVERLOAD_CHECK
239#include "clang/Basic/arm_neon.inc"
240#undef GET_NEON_OVERLOAD_CHECK
241  }
242
243  // For NEON intrinsics which are overloaded on vector element type, validate
244  // the immediate which specifies which variant to emit.
245  if (mask) {
246    unsigned ArgNo = TheCall->getNumArgs()-1;
247    if (SemaBuiltinConstantArg(TheCall, ArgNo, Result))
248      return true;
249
250    TV = Result.getLimitedValue(32);
251    if ((TV > 31) || (mask & (1 << TV)) == 0)
252      return Diag(TheCall->getLocStart(), diag::err_invalid_neon_type_code)
253        << TheCall->getArg(ArgNo)->getSourceRange();
254  }
255
256  // For NEON intrinsics which take an immediate value as part of the
257  // instruction, range check them here.
258  unsigned i = 0, l = 0, u = 0;
259  switch (BuiltinID) {
260  default: return false;
261  case ARM::BI__builtin_arm_ssat: i = 1; l = 1; u = 31; break;
262  case ARM::BI__builtin_arm_usat: i = 1; u = 31; break;
263  case ARM::BI__builtin_arm_vcvtr_f:
264  case ARM::BI__builtin_arm_vcvtr_d: i = 1; u = 1; break;
265#define GET_NEON_IMMEDIATE_CHECK
266#include "clang/Basic/arm_neon.inc"
267#undef GET_NEON_IMMEDIATE_CHECK
268  };
269
270  // Check that the immediate argument is actually a constant.
271  if (SemaBuiltinConstantArg(TheCall, i, Result))
272    return true;
273
274  // Range check against the upper/lower values for this isntruction.
275  unsigned Val = Result.getZExtValue();
276  if (Val < l || Val > (u + l))
277    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
278      << l << u+l << TheCall->getArg(i)->getSourceRange();
279
280  // FIXME: VFP Intrinsics should error if VFP not present.
281  return false;
282}
283
284/// CheckFunctionCall - Check a direct function call for various correctness
285/// and safety properties not strictly enforced by the C type system.
286bool Sema::CheckFunctionCall(FunctionDecl *FDecl, CallExpr *TheCall) {
287  // Get the IdentifierInfo* for the called function.
288  IdentifierInfo *FnInfo = FDecl->getIdentifier();
289
290  // None of the checks below are needed for functions that don't have
291  // simple names (e.g., C++ conversion functions).
292  if (!FnInfo)
293    return false;
294
295  // FIXME: This mechanism should be abstracted to be less fragile and
296  // more efficient. For example, just map function ids to custom
297  // handlers.
298
299  // Printf and scanf checking.
300  for (specific_attr_iterator<FormatAttr>
301         i = FDecl->specific_attr_begin<FormatAttr>(),
302         e = FDecl->specific_attr_end<FormatAttr>(); i != e ; ++i) {
303
304    const FormatAttr *Format = *i;
305    const bool b = Format->getType() == "scanf";
306    if (b || CheckablePrintfAttr(Format, TheCall)) {
307      bool HasVAListArg = Format->getFirstArg() == 0;
308      CheckPrintfScanfArguments(TheCall, HasVAListArg,
309                                Format->getFormatIdx() - 1,
310                                HasVAListArg ? 0 : Format->getFirstArg() - 1,
311                                !b);
312    }
313  }
314
315  for (specific_attr_iterator<NonNullAttr>
316         i = FDecl->specific_attr_begin<NonNullAttr>(),
317         e = FDecl->specific_attr_end<NonNullAttr>(); i != e; ++i) {
318    CheckNonNullArguments(*i, TheCall->getArgs(),
319                          TheCall->getCallee()->getLocStart());
320  }
321
322  // Memset/memcpy/memmove handling
323  if (FDecl->getLinkage() == ExternalLinkage &&
324      (!getLangOptions().CPlusPlus || FDecl->isExternC())) {
325    if (FnInfo->isStr("memset") || FnInfo->isStr("memcpy") ||
326        FnInfo->isStr("memmove"))
327      CheckMemsetcpymoveArguments(TheCall, FnInfo);
328  }
329
330  return false;
331}
332
333bool Sema::CheckBlockCall(NamedDecl *NDecl, CallExpr *TheCall) {
334  // Printf checking.
335  const FormatAttr *Format = NDecl->getAttr<FormatAttr>();
336  if (!Format)
337    return false;
338
339  const VarDecl *V = dyn_cast<VarDecl>(NDecl);
340  if (!V)
341    return false;
342
343  QualType Ty = V->getType();
344  if (!Ty->isBlockPointerType())
345    return false;
346
347  const bool b = Format->getType() == "scanf";
348  if (!b && !CheckablePrintfAttr(Format, TheCall))
349    return false;
350
351  bool HasVAListArg = Format->getFirstArg() == 0;
352  CheckPrintfScanfArguments(TheCall, HasVAListArg, Format->getFormatIdx() - 1,
353                            HasVAListArg ? 0 : Format->getFirstArg() - 1, !b);
354
355  return false;
356}
357
358/// SemaBuiltinAtomicOverloaded - We have a call to a function like
359/// __sync_fetch_and_add, which is an overloaded function based on the pointer
360/// type of its first argument.  The main ActOnCallExpr routines have already
361/// promoted the types of arguments because all of these calls are prototyped as
362/// void(...).
363///
364/// This function goes through and does final semantic checking for these
365/// builtins,
366ExprResult
367Sema::SemaBuiltinAtomicOverloaded(ExprResult TheCallResult) {
368  CallExpr *TheCall = (CallExpr *)TheCallResult.get();
369  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
370  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
371
372  // Ensure that we have at least one argument to do type inference from.
373  if (TheCall->getNumArgs() < 1) {
374    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
375      << 0 << 1 << TheCall->getNumArgs()
376      << TheCall->getCallee()->getSourceRange();
377    return ExprError();
378  }
379
380  // Inspect the first argument of the atomic builtin.  This should always be
381  // a pointer type, whose element is an integral scalar or pointer type.
382  // Because it is a pointer type, we don't have to worry about any implicit
383  // casts here.
384  // FIXME: We don't allow floating point scalars as input.
385  Expr *FirstArg = TheCall->getArg(0);
386  const PointerType *pointerType = FirstArg->getType()->getAs<PointerType>();
387  if (!pointerType) {
388    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer)
389      << FirstArg->getType() << FirstArg->getSourceRange();
390    return ExprError();
391  }
392
393  QualType ValType = pointerType->getPointeeType();
394  if (!ValType->isIntegerType() && !ValType->isAnyPointerType() &&
395      !ValType->isBlockPointerType()) {
396    Diag(DRE->getLocStart(), diag::err_atomic_builtin_must_be_pointer_intptr)
397      << FirstArg->getType() << FirstArg->getSourceRange();
398    return ExprError();
399  }
400
401  switch (ValType.getObjCLifetime()) {
402  case Qualifiers::OCL_None:
403  case Qualifiers::OCL_ExplicitNone:
404    // okay
405    break;
406
407  case Qualifiers::OCL_Weak:
408  case Qualifiers::OCL_Strong:
409  case Qualifiers::OCL_Autoreleasing:
410    Diag(DRE->getLocStart(), diag::err_arc_atomic_lifetime)
411      << ValType << FirstArg->getSourceRange();
412    return ExprError();
413  }
414
415  // The majority of builtins return a value, but a few have special return
416  // types, so allow them to override appropriately below.
417  QualType ResultType = ValType;
418
419  // We need to figure out which concrete builtin this maps onto.  For example,
420  // __sync_fetch_and_add with a 2 byte object turns into
421  // __sync_fetch_and_add_2.
422#define BUILTIN_ROW(x) \
423  { Builtin::BI##x##_1, Builtin::BI##x##_2, Builtin::BI##x##_4, \
424    Builtin::BI##x##_8, Builtin::BI##x##_16 }
425
426  static const unsigned BuiltinIndices[][5] = {
427    BUILTIN_ROW(__sync_fetch_and_add),
428    BUILTIN_ROW(__sync_fetch_and_sub),
429    BUILTIN_ROW(__sync_fetch_and_or),
430    BUILTIN_ROW(__sync_fetch_and_and),
431    BUILTIN_ROW(__sync_fetch_and_xor),
432
433    BUILTIN_ROW(__sync_add_and_fetch),
434    BUILTIN_ROW(__sync_sub_and_fetch),
435    BUILTIN_ROW(__sync_and_and_fetch),
436    BUILTIN_ROW(__sync_or_and_fetch),
437    BUILTIN_ROW(__sync_xor_and_fetch),
438
439    BUILTIN_ROW(__sync_val_compare_and_swap),
440    BUILTIN_ROW(__sync_bool_compare_and_swap),
441    BUILTIN_ROW(__sync_lock_test_and_set),
442    BUILTIN_ROW(__sync_lock_release),
443    BUILTIN_ROW(__sync_swap)
444  };
445#undef BUILTIN_ROW
446
447  // Determine the index of the size.
448  unsigned SizeIndex;
449  switch (Context.getTypeSizeInChars(ValType).getQuantity()) {
450  case 1: SizeIndex = 0; break;
451  case 2: SizeIndex = 1; break;
452  case 4: SizeIndex = 2; break;
453  case 8: SizeIndex = 3; break;
454  case 16: SizeIndex = 4; break;
455  default:
456    Diag(DRE->getLocStart(), diag::err_atomic_builtin_pointer_size)
457      << FirstArg->getType() << FirstArg->getSourceRange();
458    return ExprError();
459  }
460
461  // Each of these builtins has one pointer argument, followed by some number of
462  // values (0, 1 or 2) followed by a potentially empty varags list of stuff
463  // that we ignore.  Find out which row of BuiltinIndices to read from as well
464  // as the number of fixed args.
465  unsigned BuiltinID = FDecl->getBuiltinID();
466  unsigned BuiltinIndex, NumFixed = 1;
467  switch (BuiltinID) {
468  default: assert(0 && "Unknown overloaded atomic builtin!");
469  case Builtin::BI__sync_fetch_and_add: BuiltinIndex = 0; break;
470  case Builtin::BI__sync_fetch_and_sub: BuiltinIndex = 1; break;
471  case Builtin::BI__sync_fetch_and_or:  BuiltinIndex = 2; break;
472  case Builtin::BI__sync_fetch_and_and: BuiltinIndex = 3; break;
473  case Builtin::BI__sync_fetch_and_xor: BuiltinIndex = 4; break;
474
475  case Builtin::BI__sync_add_and_fetch: BuiltinIndex = 5; break;
476  case Builtin::BI__sync_sub_and_fetch: BuiltinIndex = 6; break;
477  case Builtin::BI__sync_and_and_fetch: BuiltinIndex = 7; break;
478  case Builtin::BI__sync_or_and_fetch:  BuiltinIndex = 8; break;
479  case Builtin::BI__sync_xor_and_fetch: BuiltinIndex = 9; break;
480
481  case Builtin::BI__sync_val_compare_and_swap:
482    BuiltinIndex = 10;
483    NumFixed = 2;
484    break;
485  case Builtin::BI__sync_bool_compare_and_swap:
486    BuiltinIndex = 11;
487    NumFixed = 2;
488    ResultType = Context.BoolTy;
489    break;
490  case Builtin::BI__sync_lock_test_and_set: BuiltinIndex = 12; break;
491  case Builtin::BI__sync_lock_release:
492    BuiltinIndex = 13;
493    NumFixed = 0;
494    ResultType = Context.VoidTy;
495    break;
496  case Builtin::BI__sync_swap: BuiltinIndex = 14; break;
497  }
498
499  // Now that we know how many fixed arguments we expect, first check that we
500  // have at least that many.
501  if (TheCall->getNumArgs() < 1+NumFixed) {
502    Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args_at_least)
503      << 0 << 1+NumFixed << TheCall->getNumArgs()
504      << TheCall->getCallee()->getSourceRange();
505    return ExprError();
506  }
507
508  // Get the decl for the concrete builtin from this, we can tell what the
509  // concrete integer type we should convert to is.
510  unsigned NewBuiltinID = BuiltinIndices[BuiltinIndex][SizeIndex];
511  const char *NewBuiltinName = Context.BuiltinInfo.GetName(NewBuiltinID);
512  IdentifierInfo *NewBuiltinII = PP.getIdentifierInfo(NewBuiltinName);
513  FunctionDecl *NewBuiltinDecl =
514    cast<FunctionDecl>(LazilyCreateBuiltin(NewBuiltinII, NewBuiltinID,
515                                           TUScope, false, DRE->getLocStart()));
516
517  // The first argument --- the pointer --- has a fixed type; we
518  // deduce the types of the rest of the arguments accordingly.  Walk
519  // the remaining arguments, converting them to the deduced value type.
520  for (unsigned i = 0; i != NumFixed; ++i) {
521    ExprResult Arg = TheCall->getArg(i+1);
522
523    // If the argument is an implicit cast, then there was a promotion due to
524    // "...", just remove it now.
525    if (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Arg.get())) {
526      Arg = ICE->getSubExpr();
527      ICE->setSubExpr(0);
528      TheCall->setArg(i+1, Arg.get());
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    CastKind Kind = CK_Invalid;
534    ExprValueKind VK = VK_RValue;
535    CXXCastPath BasePath;
536    Arg = CheckCastTypes(Arg.get()->getLocStart(), Arg.get()->getSourceRange(),
537                         ValType, Arg.take(), Kind, VK, BasePath);
538    if (Arg.isInvalid())
539      return ExprError();
540
541    // Okay, we have something that *can* be converted to the right type.  Check
542    // to see if there is a potentially weird extension going on here.  This can
543    // happen when you do an atomic operation on something like an char* and
544    // pass in 42.  The 42 gets converted to char.  This is even more strange
545    // for things like 45.123 -> char, etc.
546    // FIXME: Do this check.
547    Arg = ImpCastExprToType(Arg.take(), ValType, Kind, VK, &BasePath);
548    TheCall->setArg(i+1, Arg.get());
549  }
550
551  // Switch the DeclRefExpr to refer to the new decl.
552  DRE->setDecl(NewBuiltinDecl);
553  DRE->setType(NewBuiltinDecl->getType());
554
555  // Set the callee in the CallExpr.
556  // FIXME: This leaks the original parens and implicit casts.
557  ExprResult PromotedCall = UsualUnaryConversions(DRE);
558  if (PromotedCall.isInvalid())
559    return ExprError();
560  TheCall->setCallee(PromotedCall.take());
561
562  // Change the result type of the call to match the original value type. This
563  // is arbitrary, but the codegen for these builtins ins design to handle it
564  // gracefully.
565  TheCall->setType(ResultType);
566
567  return move(TheCallResult);
568}
569
570
571/// CheckObjCString - Checks that the argument to the builtin
572/// CFString constructor is correct
573/// Note: It might also make sense to do the UTF-16 conversion here (would
574/// simplify the backend).
575bool Sema::CheckObjCString(Expr *Arg) {
576  Arg = Arg->IgnoreParenCasts();
577  StringLiteral *Literal = dyn_cast<StringLiteral>(Arg);
578
579  if (!Literal || Literal->isWide()) {
580    Diag(Arg->getLocStart(), diag::err_cfstring_literal_not_string_constant)
581      << Arg->getSourceRange();
582    return true;
583  }
584
585  if (Literal->containsNonAsciiOrNull()) {
586    llvm::StringRef String = Literal->getString();
587    unsigned NumBytes = String.size();
588    llvm::SmallVector<UTF16, 128> ToBuf(NumBytes);
589    const UTF8 *FromPtr = (UTF8 *)String.data();
590    UTF16 *ToPtr = &ToBuf[0];
591
592    ConversionResult Result = ConvertUTF8toUTF16(&FromPtr, FromPtr + NumBytes,
593                                                 &ToPtr, ToPtr + NumBytes,
594                                                 strictConversion);
595    // Check for conversion failure.
596    if (Result != conversionOK)
597      Diag(Arg->getLocStart(),
598           diag::warn_cfstring_truncated) << Arg->getSourceRange();
599  }
600  return false;
601}
602
603/// SemaBuiltinVAStart - Check the arguments to __builtin_va_start for validity.
604/// Emit an error and return true on failure, return false on success.
605bool Sema::SemaBuiltinVAStart(CallExpr *TheCall) {
606  Expr *Fn = TheCall->getCallee();
607  if (TheCall->getNumArgs() > 2) {
608    Diag(TheCall->getArg(2)->getLocStart(),
609         diag::err_typecheck_call_too_many_args)
610      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
611      << Fn->getSourceRange()
612      << SourceRange(TheCall->getArg(2)->getLocStart(),
613                     (*(TheCall->arg_end()-1))->getLocEnd());
614    return true;
615  }
616
617  if (TheCall->getNumArgs() < 2) {
618    return Diag(TheCall->getLocEnd(),
619      diag::err_typecheck_call_too_few_args_at_least)
620      << 0 /*function call*/ << 2 << TheCall->getNumArgs();
621  }
622
623  // Determine whether the current function is variadic or not.
624  BlockScopeInfo *CurBlock = getCurBlock();
625  bool isVariadic;
626  if (CurBlock)
627    isVariadic = CurBlock->TheDecl->isVariadic();
628  else if (FunctionDecl *FD = getCurFunctionDecl())
629    isVariadic = FD->isVariadic();
630  else
631    isVariadic = getCurMethodDecl()->isVariadic();
632
633  if (!isVariadic) {
634    Diag(Fn->getLocStart(), diag::err_va_start_used_in_non_variadic_function);
635    return true;
636  }
637
638  // Verify that the second argument to the builtin is the last argument of the
639  // current function or method.
640  bool SecondArgIsLastNamedArgument = false;
641  const Expr *Arg = TheCall->getArg(1)->IgnoreParenCasts();
642
643  if (const DeclRefExpr *DR = dyn_cast<DeclRefExpr>(Arg)) {
644    if (const ParmVarDecl *PV = dyn_cast<ParmVarDecl>(DR->getDecl())) {
645      // FIXME: This isn't correct for methods (results in bogus warning).
646      // Get the last formal in the current function.
647      const ParmVarDecl *LastArg;
648      if (CurBlock)
649        LastArg = *(CurBlock->TheDecl->param_end()-1);
650      else if (FunctionDecl *FD = getCurFunctionDecl())
651        LastArg = *(FD->param_end()-1);
652      else
653        LastArg = *(getCurMethodDecl()->param_end()-1);
654      SecondArgIsLastNamedArgument = PV == LastArg;
655    }
656  }
657
658  if (!SecondArgIsLastNamedArgument)
659    Diag(TheCall->getArg(1)->getLocStart(),
660         diag::warn_second_parameter_of_va_start_not_last_named_argument);
661  return false;
662}
663
664/// SemaBuiltinUnorderedCompare - Handle functions like __builtin_isgreater and
665/// friends.  This is declared to take (...), so we have to check everything.
666bool Sema::SemaBuiltinUnorderedCompare(CallExpr *TheCall) {
667  if (TheCall->getNumArgs() < 2)
668    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
669      << 0 << 2 << TheCall->getNumArgs()/*function call*/;
670  if (TheCall->getNumArgs() > 2)
671    return Diag(TheCall->getArg(2)->getLocStart(),
672                diag::err_typecheck_call_too_many_args)
673      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
674      << SourceRange(TheCall->getArg(2)->getLocStart(),
675                     (*(TheCall->arg_end()-1))->getLocEnd());
676
677  ExprResult OrigArg0 = TheCall->getArg(0);
678  ExprResult OrigArg1 = TheCall->getArg(1);
679
680  // Do standard promotions between the two arguments, returning their common
681  // type.
682  QualType Res = UsualArithmeticConversions(OrigArg0, OrigArg1, false);
683  if (OrigArg0.isInvalid() || OrigArg1.isInvalid())
684    return true;
685
686  // Make sure any conversions are pushed back into the call; this is
687  // type safe since unordered compare builtins are declared as "_Bool
688  // foo(...)".
689  TheCall->setArg(0, OrigArg0.get());
690  TheCall->setArg(1, OrigArg1.get());
691
692  if (OrigArg0.get()->isTypeDependent() || OrigArg1.get()->isTypeDependent())
693    return false;
694
695  // If the common type isn't a real floating type, then the arguments were
696  // invalid for this operation.
697  if (!Res->isRealFloatingType())
698    return Diag(OrigArg0.get()->getLocStart(),
699                diag::err_typecheck_call_invalid_ordered_compare)
700      << OrigArg0.get()->getType() << OrigArg1.get()->getType()
701      << SourceRange(OrigArg0.get()->getLocStart(), OrigArg1.get()->getLocEnd());
702
703  return false;
704}
705
706/// SemaBuiltinSemaBuiltinFPClassification - Handle functions like
707/// __builtin_isnan and friends.  This is declared to take (...), so we have
708/// to check everything. We expect the last argument to be a floating point
709/// value.
710bool Sema::SemaBuiltinFPClassification(CallExpr *TheCall, unsigned NumArgs) {
711  if (TheCall->getNumArgs() < NumArgs)
712    return Diag(TheCall->getLocEnd(), diag::err_typecheck_call_too_few_args)
713      << 0 << NumArgs << TheCall->getNumArgs()/*function call*/;
714  if (TheCall->getNumArgs() > NumArgs)
715    return Diag(TheCall->getArg(NumArgs)->getLocStart(),
716                diag::err_typecheck_call_too_many_args)
717      << 0 /*function call*/ << NumArgs << TheCall->getNumArgs()
718      << SourceRange(TheCall->getArg(NumArgs)->getLocStart(),
719                     (*(TheCall->arg_end()-1))->getLocEnd());
720
721  Expr *OrigArg = TheCall->getArg(NumArgs-1);
722
723  if (OrigArg->isTypeDependent())
724    return false;
725
726  // This operation requires a non-_Complex floating-point number.
727  if (!OrigArg->getType()->isRealFloatingType())
728    return Diag(OrigArg->getLocStart(),
729                diag::err_typecheck_call_invalid_unary_fp)
730      << OrigArg->getType() << OrigArg->getSourceRange();
731
732  // If this is an implicit conversion from float -> double, remove it.
733  if (ImplicitCastExpr *Cast = dyn_cast<ImplicitCastExpr>(OrigArg)) {
734    Expr *CastArg = Cast->getSubExpr();
735    if (CastArg->getType()->isSpecificBuiltinType(BuiltinType::Float)) {
736      assert(Cast->getType()->isSpecificBuiltinType(BuiltinType::Double) &&
737             "promotion from float to double is the only expected cast here");
738      Cast->setSubExpr(0);
739      TheCall->setArg(NumArgs-1, CastArg);
740      OrigArg = CastArg;
741    }
742  }
743
744  return false;
745}
746
747/// SemaBuiltinShuffleVector - Handle __builtin_shufflevector.
748// This is declared to take (...), so we have to check everything.
749ExprResult Sema::SemaBuiltinShuffleVector(CallExpr *TheCall) {
750  if (TheCall->getNumArgs() < 2)
751    return ExprError(Diag(TheCall->getLocEnd(),
752                          diag::err_typecheck_call_too_few_args_at_least)
753      << 0 /*function call*/ << 2 << TheCall->getNumArgs()
754      << TheCall->getSourceRange());
755
756  // Determine which of the following types of shufflevector we're checking:
757  // 1) unary, vector mask: (lhs, mask)
758  // 2) binary, vector mask: (lhs, rhs, mask)
759  // 3) binary, scalar mask: (lhs, rhs, index, ..., index)
760  QualType resType = TheCall->getArg(0)->getType();
761  unsigned numElements = 0;
762
763  if (!TheCall->getArg(0)->isTypeDependent() &&
764      !TheCall->getArg(1)->isTypeDependent()) {
765    QualType LHSType = TheCall->getArg(0)->getType();
766    QualType RHSType = TheCall->getArg(1)->getType();
767
768    if (!LHSType->isVectorType() || !RHSType->isVectorType()) {
769      Diag(TheCall->getLocStart(), diag::err_shufflevector_non_vector)
770        << SourceRange(TheCall->getArg(0)->getLocStart(),
771                       TheCall->getArg(1)->getLocEnd());
772      return ExprError();
773    }
774
775    numElements = LHSType->getAs<VectorType>()->getNumElements();
776    unsigned numResElements = TheCall->getNumArgs() - 2;
777
778    // Check to see if we have a call with 2 vector arguments, the unary shuffle
779    // with mask.  If so, verify that RHS is an integer vector type with the
780    // same number of elts as lhs.
781    if (TheCall->getNumArgs() == 2) {
782      if (!RHSType->hasIntegerRepresentation() ||
783          RHSType->getAs<VectorType>()->getNumElements() != numElements)
784        Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
785          << SourceRange(TheCall->getArg(1)->getLocStart(),
786                         TheCall->getArg(1)->getLocEnd());
787      numResElements = numElements;
788    }
789    else if (!Context.hasSameUnqualifiedType(LHSType, RHSType)) {
790      Diag(TheCall->getLocStart(), diag::err_shufflevector_incompatible_vector)
791        << SourceRange(TheCall->getArg(0)->getLocStart(),
792                       TheCall->getArg(1)->getLocEnd());
793      return ExprError();
794    } else if (numElements != numResElements) {
795      QualType eltType = LHSType->getAs<VectorType>()->getElementType();
796      resType = Context.getVectorType(eltType, numResElements,
797                                      VectorType::GenericVector);
798    }
799  }
800
801  for (unsigned i = 2; i < TheCall->getNumArgs(); i++) {
802    if (TheCall->getArg(i)->isTypeDependent() ||
803        TheCall->getArg(i)->isValueDependent())
804      continue;
805
806    llvm::APSInt Result(32);
807    if (!TheCall->getArg(i)->isIntegerConstantExpr(Result, Context))
808      return ExprError(Diag(TheCall->getLocStart(),
809                  diag::err_shufflevector_nonconstant_argument)
810                << TheCall->getArg(i)->getSourceRange());
811
812    if (Result.getActiveBits() > 64 || Result.getZExtValue() >= numElements*2)
813      return ExprError(Diag(TheCall->getLocStart(),
814                  diag::err_shufflevector_argument_too_large)
815               << TheCall->getArg(i)->getSourceRange());
816  }
817
818  llvm::SmallVector<Expr*, 32> exprs;
819
820  for (unsigned i = 0, e = TheCall->getNumArgs(); i != e; i++) {
821    exprs.push_back(TheCall->getArg(i));
822    TheCall->setArg(i, 0);
823  }
824
825  return Owned(new (Context) ShuffleVectorExpr(Context, exprs.begin(),
826                                            exprs.size(), resType,
827                                            TheCall->getCallee()->getLocStart(),
828                                            TheCall->getRParenLoc()));
829}
830
831/// SemaBuiltinPrefetch - Handle __builtin_prefetch.
832// This is declared to take (const void*, ...) and can take two
833// optional constant int args.
834bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
835  unsigned NumArgs = TheCall->getNumArgs();
836
837  if (NumArgs > 3)
838    return Diag(TheCall->getLocEnd(),
839             diag::err_typecheck_call_too_many_args_at_most)
840             << 0 /*function call*/ << 3 << NumArgs
841             << TheCall->getSourceRange();
842
843  // Argument 0 is checked for us and the remaining arguments must be
844  // constant integers.
845  for (unsigned i = 1; i != NumArgs; ++i) {
846    Expr *Arg = TheCall->getArg(i);
847
848    llvm::APSInt Result;
849    if (SemaBuiltinConstantArg(TheCall, i, Result))
850      return true;
851
852    // FIXME: gcc issues a warning and rewrites these to 0. These
853    // seems especially odd for the third argument since the default
854    // is 3.
855    if (i == 1) {
856      if (Result.getLimitedValue() > 1)
857        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
858             << "0" << "1" << Arg->getSourceRange();
859    } else {
860      if (Result.getLimitedValue() > 3)
861        return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
862            << "0" << "3" << Arg->getSourceRange();
863    }
864  }
865
866  return false;
867}
868
869/// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr
870/// TheCall is a constant expression.
871bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,
872                                  llvm::APSInt &Result) {
873  Expr *Arg = TheCall->getArg(ArgNum);
874  DeclRefExpr *DRE =cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
875  FunctionDecl *FDecl = cast<FunctionDecl>(DRE->getDecl());
876
877  if (Arg->isTypeDependent() || Arg->isValueDependent()) return false;
878
879  if (!Arg->isIntegerConstantExpr(Result, Context))
880    return Diag(TheCall->getLocStart(), diag::err_constant_integer_arg_type)
881                << FDecl->getDeclName() <<  Arg->getSourceRange();
882
883  return false;
884}
885
886/// SemaBuiltinObjectSize - Handle __builtin_object_size(void *ptr,
887/// int type). This simply type checks that type is one of the defined
888/// constants (0-3).
889// For compatibility check 0-3, llvm only handles 0 and 2.
890bool Sema::SemaBuiltinObjectSize(CallExpr *TheCall) {
891  llvm::APSInt Result;
892
893  // Check constant-ness first.
894  if (SemaBuiltinConstantArg(TheCall, 1, Result))
895    return true;
896
897  Expr *Arg = TheCall->getArg(1);
898  if (Result.getSExtValue() < 0 || Result.getSExtValue() > 3) {
899    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
900             << "0" << "3" << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
901  }
902
903  return false;
904}
905
906/// SemaBuiltinLongjmp - Handle __builtin_longjmp(void *env[5], int val).
907/// This checks that val is a constant 1.
908bool Sema::SemaBuiltinLongjmp(CallExpr *TheCall) {
909  Expr *Arg = TheCall->getArg(1);
910  llvm::APSInt Result;
911
912  // TODO: This is less than ideal. Overload this to take a value.
913  if (SemaBuiltinConstantArg(TheCall, 1, Result))
914    return true;
915
916  if (Result != 1)
917    return Diag(TheCall->getLocStart(), diag::err_builtin_longjmp_invalid_val)
918             << SourceRange(Arg->getLocStart(), Arg->getLocEnd());
919
920  return false;
921}
922
923// Handle i > 1 ? "x" : "y", recursively.
924bool Sema::SemaCheckStringLiteral(const Expr *E, const CallExpr *TheCall,
925                                  bool HasVAListArg,
926                                  unsigned format_idx, unsigned firstDataArg,
927                                  bool isPrintf) {
928 tryAgain:
929  if (E->isTypeDependent() || E->isValueDependent())
930    return false;
931
932  E = E->IgnoreParens();
933
934  switch (E->getStmtClass()) {
935  case Stmt::BinaryConditionalOperatorClass:
936  case Stmt::ConditionalOperatorClass: {
937    const AbstractConditionalOperator *C = cast<AbstractConditionalOperator>(E);
938    return SemaCheckStringLiteral(C->getTrueExpr(), TheCall, HasVAListArg,
939                                  format_idx, firstDataArg, isPrintf)
940        && SemaCheckStringLiteral(C->getFalseExpr(), TheCall, HasVAListArg,
941                                  format_idx, firstDataArg, isPrintf);
942  }
943
944  case Stmt::IntegerLiteralClass:
945    // Technically -Wformat-nonliteral does not warn about this case.
946    // The behavior of printf and friends in this case is implementation
947    // dependent.  Ideally if the format string cannot be null then
948    // it should have a 'nonnull' attribute in the function prototype.
949    return true;
950
951  case Stmt::ImplicitCastExprClass: {
952    E = cast<ImplicitCastExpr>(E)->getSubExpr();
953    goto tryAgain;
954  }
955
956  case Stmt::OpaqueValueExprClass:
957    if (const Expr *src = cast<OpaqueValueExpr>(E)->getSourceExpr()) {
958      E = src;
959      goto tryAgain;
960    }
961    return false;
962
963  case Stmt::PredefinedExprClass:
964    // While __func__, etc., are technically not string literals, they
965    // cannot contain format specifiers and thus are not a security
966    // liability.
967    return true;
968
969  case Stmt::DeclRefExprClass: {
970    const DeclRefExpr *DR = cast<DeclRefExpr>(E);
971
972    // As an exception, do not flag errors for variables binding to
973    // const string literals.
974    if (const VarDecl *VD = dyn_cast<VarDecl>(DR->getDecl())) {
975      bool isConstant = false;
976      QualType T = DR->getType();
977
978      if (const ArrayType *AT = Context.getAsArrayType(T)) {
979        isConstant = AT->getElementType().isConstant(Context);
980      } else if (const PointerType *PT = T->getAs<PointerType>()) {
981        isConstant = T.isConstant(Context) &&
982                     PT->getPointeeType().isConstant(Context);
983      }
984
985      if (isConstant) {
986        if (const Expr *Init = VD->getAnyInitializer())
987          return SemaCheckStringLiteral(Init, TheCall,
988                                        HasVAListArg, format_idx, firstDataArg,
989                                        isPrintf);
990      }
991
992      // For vprintf* functions (i.e., HasVAListArg==true), we add a
993      // special check to see if the format string is a function parameter
994      // of the function calling the printf function.  If the function
995      // has an attribute indicating it is a printf-like function, then we
996      // should suppress warnings concerning non-literals being used in a call
997      // to a vprintf function.  For example:
998      //
999      // void
1000      // logmessage(char const *fmt __attribute__ (format (printf, 1, 2)), ...){
1001      //      va_list ap;
1002      //      va_start(ap, fmt);
1003      //      vprintf(fmt, ap);  // Do NOT emit a warning about "fmt".
1004      //      ...
1005      //
1006      //
1007      //  FIXME: We don't have full attribute support yet, so just check to see
1008      //    if the argument is a DeclRefExpr that references a parameter.  We'll
1009      //    add proper support for checking the attribute later.
1010      if (HasVAListArg)
1011        if (isa<ParmVarDecl>(VD))
1012          return true;
1013    }
1014
1015    return false;
1016  }
1017
1018  case Stmt::CallExprClass: {
1019    const CallExpr *CE = cast<CallExpr>(E);
1020    if (const ImplicitCastExpr *ICE
1021          = dyn_cast<ImplicitCastExpr>(CE->getCallee())) {
1022      if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(ICE->getSubExpr())) {
1023        if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(DRE->getDecl())) {
1024          if (const FormatArgAttr *FA = FD->getAttr<FormatArgAttr>()) {
1025            unsigned ArgIndex = FA->getFormatIdx();
1026            const Expr *Arg = CE->getArg(ArgIndex - 1);
1027
1028            return SemaCheckStringLiteral(Arg, TheCall, HasVAListArg,
1029                                          format_idx, firstDataArg, isPrintf);
1030          }
1031        }
1032      }
1033    }
1034
1035    return false;
1036  }
1037  case Stmt::ObjCStringLiteralClass:
1038  case Stmt::StringLiteralClass: {
1039    const StringLiteral *StrE = NULL;
1040
1041    if (const ObjCStringLiteral *ObjCFExpr = dyn_cast<ObjCStringLiteral>(E))
1042      StrE = ObjCFExpr->getString();
1043    else
1044      StrE = cast<StringLiteral>(E);
1045
1046    if (StrE) {
1047      CheckFormatString(StrE, E, TheCall, HasVAListArg, format_idx,
1048                        firstDataArg, isPrintf);
1049      return true;
1050    }
1051
1052    return false;
1053  }
1054
1055  default:
1056    return false;
1057  }
1058}
1059
1060void
1061Sema::CheckNonNullArguments(const NonNullAttr *NonNull,
1062                            const Expr * const *ExprArgs,
1063                            SourceLocation CallSiteLoc) {
1064  for (NonNullAttr::args_iterator i = NonNull->args_begin(),
1065                                  e = NonNull->args_end();
1066       i != e; ++i) {
1067    const Expr *ArgExpr = ExprArgs[*i];
1068    if (ArgExpr->isNullPointerConstant(Context,
1069                                       Expr::NPC_ValueDependentIsNotNull))
1070      Diag(CallSiteLoc, diag::warn_null_arg) << ArgExpr->getSourceRange();
1071  }
1072}
1073
1074/// CheckPrintfScanfArguments - Check calls to printf and scanf (and similar
1075/// functions) for correct use of format strings.
1076void
1077Sema::CheckPrintfScanfArguments(const CallExpr *TheCall, bool HasVAListArg,
1078                                unsigned format_idx, unsigned firstDataArg,
1079                                bool isPrintf) {
1080
1081  const Expr *Fn = TheCall->getCallee();
1082
1083  // The way the format attribute works in GCC, the implicit this argument
1084  // of member functions is counted. However, it doesn't appear in our own
1085  // lists, so decrement format_idx in that case.
1086  if (isa<CXXMemberCallExpr>(TheCall)) {
1087    const CXXMethodDecl *method_decl =
1088      dyn_cast<CXXMethodDecl>(TheCall->getCalleeDecl());
1089    if (method_decl && method_decl->isInstance()) {
1090      // Catch a format attribute mistakenly referring to the object argument.
1091      if (format_idx == 0)
1092        return;
1093      --format_idx;
1094      if(firstDataArg != 0)
1095        --firstDataArg;
1096    }
1097  }
1098
1099  // CHECK: printf/scanf-like function is called with no format string.
1100  if (format_idx >= TheCall->getNumArgs()) {
1101    Diag(TheCall->getRParenLoc(), diag::warn_missing_format_string)
1102      << Fn->getSourceRange();
1103    return;
1104  }
1105
1106  const Expr *OrigFormatExpr = TheCall->getArg(format_idx)->IgnoreParenCasts();
1107
1108  // CHECK: format string is not a string literal.
1109  //
1110  // Dynamically generated format strings are difficult to
1111  // automatically vet at compile time.  Requiring that format strings
1112  // are string literals: (1) permits the checking of format strings by
1113  // the compiler and thereby (2) can practically remove the source of
1114  // many format string exploits.
1115
1116  // Format string can be either ObjC string (e.g. @"%d") or
1117  // C string (e.g. "%d")
1118  // ObjC string uses the same format specifiers as C string, so we can use
1119  // the same format string checking logic for both ObjC and C strings.
1120  if (SemaCheckStringLiteral(OrigFormatExpr, TheCall, HasVAListArg, format_idx,
1121                             firstDataArg, isPrintf))
1122    return;  // Literal format string found, check done!
1123
1124  // If there are no arguments specified, warn with -Wformat-security, otherwise
1125  // warn only with -Wformat-nonliteral.
1126  if (TheCall->getNumArgs() == format_idx+1)
1127    Diag(TheCall->getArg(format_idx)->getLocStart(),
1128         diag::warn_format_nonliteral_noargs)
1129      << OrigFormatExpr->getSourceRange();
1130  else
1131    Diag(TheCall->getArg(format_idx)->getLocStart(),
1132         diag::warn_format_nonliteral)
1133           << OrigFormatExpr->getSourceRange();
1134}
1135
1136namespace {
1137class CheckFormatHandler : public analyze_format_string::FormatStringHandler {
1138protected:
1139  Sema &S;
1140  const StringLiteral *FExpr;
1141  const Expr *OrigFormatExpr;
1142  const unsigned FirstDataArg;
1143  const unsigned NumDataArgs;
1144  const bool IsObjCLiteral;
1145  const char *Beg; // Start of format string.
1146  const bool HasVAListArg;
1147  const CallExpr *TheCall;
1148  unsigned FormatIdx;
1149  llvm::BitVector CoveredArgs;
1150  bool usesPositionalArgs;
1151  bool atFirstArg;
1152public:
1153  CheckFormatHandler(Sema &s, const StringLiteral *fexpr,
1154                     const Expr *origFormatExpr, unsigned firstDataArg,
1155                     unsigned numDataArgs, bool isObjCLiteral,
1156                     const char *beg, bool hasVAListArg,
1157                     const CallExpr *theCall, unsigned formatIdx)
1158    : S(s), FExpr(fexpr), OrigFormatExpr(origFormatExpr),
1159      FirstDataArg(firstDataArg),
1160      NumDataArgs(numDataArgs),
1161      IsObjCLiteral(isObjCLiteral), Beg(beg),
1162      HasVAListArg(hasVAListArg),
1163      TheCall(theCall), FormatIdx(formatIdx),
1164      usesPositionalArgs(false), atFirstArg(true) {
1165        CoveredArgs.resize(numDataArgs);
1166        CoveredArgs.reset();
1167      }
1168
1169  void DoneProcessing();
1170
1171  void HandleIncompleteSpecifier(const char *startSpecifier,
1172                                 unsigned specifierLen);
1173
1174  virtual void HandleInvalidPosition(const char *startSpecifier,
1175                                     unsigned specifierLen,
1176                                     analyze_format_string::PositionContext p);
1177
1178  virtual void HandleZeroPosition(const char *startPos, unsigned posLen);
1179
1180  void HandleNullChar(const char *nullCharacter);
1181
1182protected:
1183  bool HandleInvalidConversionSpecifier(unsigned argIndex, SourceLocation Loc,
1184                                        const char *startSpec,
1185                                        unsigned specifierLen,
1186                                        const char *csStart, unsigned csLen);
1187
1188  SourceRange getFormatStringRange();
1189  CharSourceRange getSpecifierRange(const char *startSpecifier,
1190                                    unsigned specifierLen);
1191  SourceLocation getLocationOfByte(const char *x);
1192
1193  const Expr *getDataArg(unsigned i) const;
1194
1195  bool CheckNumArgs(const analyze_format_string::FormatSpecifier &FS,
1196                    const analyze_format_string::ConversionSpecifier &CS,
1197                    const char *startSpecifier, unsigned specifierLen,
1198                    unsigned argIndex);
1199};
1200}
1201
1202SourceRange CheckFormatHandler::getFormatStringRange() {
1203  return OrigFormatExpr->getSourceRange();
1204}
1205
1206CharSourceRange CheckFormatHandler::
1207getSpecifierRange(const char *startSpecifier, unsigned specifierLen) {
1208  SourceLocation Start = getLocationOfByte(startSpecifier);
1209  SourceLocation End   = getLocationOfByte(startSpecifier + specifierLen - 1);
1210
1211  // Advance the end SourceLocation by one due to half-open ranges.
1212  End = End.getFileLocWithOffset(1);
1213
1214  return CharSourceRange::getCharRange(Start, End);
1215}
1216
1217SourceLocation CheckFormatHandler::getLocationOfByte(const char *x) {
1218  return S.getLocationOfStringLiteralByte(FExpr, x - Beg);
1219}
1220
1221void CheckFormatHandler::HandleIncompleteSpecifier(const char *startSpecifier,
1222                                                   unsigned specifierLen){
1223  SourceLocation Loc = getLocationOfByte(startSpecifier);
1224  S.Diag(Loc, diag::warn_printf_incomplete_specifier)
1225    << getSpecifierRange(startSpecifier, specifierLen);
1226}
1227
1228void
1229CheckFormatHandler::HandleInvalidPosition(const char *startPos, unsigned posLen,
1230                                     analyze_format_string::PositionContext p) {
1231  SourceLocation Loc = getLocationOfByte(startPos);
1232  S.Diag(Loc, diag::warn_format_invalid_positional_specifier)
1233    << (unsigned) p << getSpecifierRange(startPos, posLen);
1234}
1235
1236void CheckFormatHandler::HandleZeroPosition(const char *startPos,
1237                                            unsigned posLen) {
1238  SourceLocation Loc = getLocationOfByte(startPos);
1239  S.Diag(Loc, diag::warn_format_zero_positional_specifier)
1240    << getSpecifierRange(startPos, posLen);
1241}
1242
1243void CheckFormatHandler::HandleNullChar(const char *nullCharacter) {
1244  if (!IsObjCLiteral) {
1245    // The presence of a null character is likely an error.
1246    S.Diag(getLocationOfByte(nullCharacter),
1247           diag::warn_printf_format_string_contains_null_char)
1248      << getFormatStringRange();
1249  }
1250}
1251
1252const Expr *CheckFormatHandler::getDataArg(unsigned i) const {
1253  return TheCall->getArg(FirstDataArg + i);
1254}
1255
1256void CheckFormatHandler::DoneProcessing() {
1257    // Does the number of data arguments exceed the number of
1258    // format conversions in the format string?
1259  if (!HasVAListArg) {
1260      // Find any arguments that weren't covered.
1261    CoveredArgs.flip();
1262    signed notCoveredArg = CoveredArgs.find_first();
1263    if (notCoveredArg >= 0) {
1264      assert((unsigned)notCoveredArg < NumDataArgs);
1265      S.Diag(getDataArg((unsigned) notCoveredArg)->getLocStart(),
1266             diag::warn_printf_data_arg_not_used)
1267      << getFormatStringRange();
1268    }
1269  }
1270}
1271
1272bool
1273CheckFormatHandler::HandleInvalidConversionSpecifier(unsigned argIndex,
1274                                                     SourceLocation Loc,
1275                                                     const char *startSpec,
1276                                                     unsigned specifierLen,
1277                                                     const char *csStart,
1278                                                     unsigned csLen) {
1279
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  S.Diag(Loc, diag::warn_format_invalid_conversion)
1296    << llvm::StringRef(csStart, csLen)
1297    << getSpecifierRange(startSpec, specifierLen);
1298
1299  return keepGoing;
1300}
1301
1302bool
1303CheckFormatHandler::CheckNumArgs(
1304  const analyze_format_string::FormatSpecifier &FS,
1305  const analyze_format_string::ConversionSpecifier &CS,
1306  const char *startSpecifier, unsigned specifierLen, unsigned argIndex) {
1307
1308  if (argIndex >= NumDataArgs) {
1309    if (FS.usesPositionalArg())  {
1310      S.Diag(getLocationOfByte(CS.getStart()),
1311             diag::warn_printf_positional_arg_exceeds_data_args)
1312      << (argIndex+1) << NumDataArgs
1313      << getSpecifierRange(startSpecifier, specifierLen);
1314    }
1315    else {
1316      S.Diag(getLocationOfByte(CS.getStart()),
1317             diag::warn_printf_insufficient_data_args)
1318      << getSpecifierRange(startSpecifier, specifierLen);
1319    }
1320
1321    return false;
1322  }
1323  return true;
1324}
1325
1326//===--- CHECK: Printf format string checking ------------------------------===//
1327
1328namespace {
1329class CheckPrintfHandler : public CheckFormatHandler {
1330public:
1331  CheckPrintfHandler(Sema &s, const StringLiteral *fexpr,
1332                     const Expr *origFormatExpr, unsigned firstDataArg,
1333                     unsigned numDataArgs, bool isObjCLiteral,
1334                     const char *beg, bool hasVAListArg,
1335                     const CallExpr *theCall, unsigned formatIdx)
1336  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1337                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1338                       theCall, formatIdx) {}
1339
1340
1341  bool HandleInvalidPrintfConversionSpecifier(
1342                                      const analyze_printf::PrintfSpecifier &FS,
1343                                      const char *startSpecifier,
1344                                      unsigned specifierLen);
1345
1346  bool HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier &FS,
1347                             const char *startSpecifier,
1348                             unsigned specifierLen);
1349
1350  bool HandleAmount(const analyze_format_string::OptionalAmount &Amt, unsigned k,
1351                    const char *startSpecifier, unsigned specifierLen);
1352  void HandleInvalidAmount(const analyze_printf::PrintfSpecifier &FS,
1353                           const analyze_printf::OptionalAmount &Amt,
1354                           unsigned type,
1355                           const char *startSpecifier, unsigned specifierLen);
1356  void HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1357                  const analyze_printf::OptionalFlag &flag,
1358                  const char *startSpecifier, unsigned specifierLen);
1359  void HandleIgnoredFlag(const analyze_printf::PrintfSpecifier &FS,
1360                         const analyze_printf::OptionalFlag &ignoredFlag,
1361                         const analyze_printf::OptionalFlag &flag,
1362                         const char *startSpecifier, unsigned specifierLen);
1363};
1364}
1365
1366bool CheckPrintfHandler::HandleInvalidPrintfConversionSpecifier(
1367                                      const analyze_printf::PrintfSpecifier &FS,
1368                                      const char *startSpecifier,
1369                                      unsigned specifierLen) {
1370  const analyze_printf::PrintfConversionSpecifier &CS =
1371    FS.getConversionSpecifier();
1372
1373  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1374                                          getLocationOfByte(CS.getStart()),
1375                                          startSpecifier, specifierLen,
1376                                          CS.getStart(), CS.getLength());
1377}
1378
1379bool CheckPrintfHandler::HandleAmount(
1380                               const analyze_format_string::OptionalAmount &Amt,
1381                               unsigned k, const char *startSpecifier,
1382                               unsigned specifierLen) {
1383
1384  if (Amt.hasDataArgument()) {
1385    if (!HasVAListArg) {
1386      unsigned argIndex = Amt.getArgIndex();
1387      if (argIndex >= NumDataArgs) {
1388        S.Diag(getLocationOfByte(Amt.getStart()),
1389               diag::warn_printf_asterisk_missing_arg)
1390          << k << getSpecifierRange(startSpecifier, specifierLen);
1391        // Don't do any more checking.  We will just emit
1392        // spurious errors.
1393        return false;
1394      }
1395
1396      // Type check the data argument.  It should be an 'int'.
1397      // Although not in conformance with C99, we also allow the argument to be
1398      // an 'unsigned int' as that is a reasonably safe case.  GCC also
1399      // doesn't emit a warning for that case.
1400      CoveredArgs.set(argIndex);
1401      const Expr *Arg = getDataArg(argIndex);
1402      QualType T = Arg->getType();
1403
1404      const analyze_printf::ArgTypeResult &ATR = Amt.getArgType(S.Context);
1405      assert(ATR.isValid());
1406
1407      if (!ATR.matchesType(S.Context, T)) {
1408        S.Diag(getLocationOfByte(Amt.getStart()),
1409               diag::warn_printf_asterisk_wrong_type)
1410          << k
1411          << ATR.getRepresentativeType(S.Context) << T
1412          << getSpecifierRange(startSpecifier, specifierLen)
1413          << Arg->getSourceRange();
1414        // Don't do any more checking.  We will just emit
1415        // spurious errors.
1416        return false;
1417      }
1418    }
1419  }
1420  return true;
1421}
1422
1423void CheckPrintfHandler::HandleInvalidAmount(
1424                                      const analyze_printf::PrintfSpecifier &FS,
1425                                      const analyze_printf::OptionalAmount &Amt,
1426                                      unsigned type,
1427                                      const char *startSpecifier,
1428                                      unsigned specifierLen) {
1429  const analyze_printf::PrintfConversionSpecifier &CS =
1430    FS.getConversionSpecifier();
1431  switch (Amt.getHowSpecified()) {
1432  case analyze_printf::OptionalAmount::Constant:
1433    S.Diag(getLocationOfByte(Amt.getStart()),
1434        diag::warn_printf_nonsensical_optional_amount)
1435      << type
1436      << CS.toString()
1437      << getSpecifierRange(startSpecifier, specifierLen)
1438      << FixItHint::CreateRemoval(getSpecifierRange(Amt.getStart(),
1439          Amt.getConstantLength()));
1440    break;
1441
1442  default:
1443    S.Diag(getLocationOfByte(Amt.getStart()),
1444        diag::warn_printf_nonsensical_optional_amount)
1445      << type
1446      << CS.toString()
1447      << getSpecifierRange(startSpecifier, specifierLen);
1448    break;
1449  }
1450}
1451
1452void CheckPrintfHandler::HandleFlag(const analyze_printf::PrintfSpecifier &FS,
1453                                    const analyze_printf::OptionalFlag &flag,
1454                                    const char *startSpecifier,
1455                                    unsigned specifierLen) {
1456  // Warn about pointless flag with a fixit removal.
1457  const analyze_printf::PrintfConversionSpecifier &CS =
1458    FS.getConversionSpecifier();
1459  S.Diag(getLocationOfByte(flag.getPosition()),
1460      diag::warn_printf_nonsensical_flag)
1461    << flag.toString() << CS.toString()
1462    << getSpecifierRange(startSpecifier, specifierLen)
1463    << FixItHint::CreateRemoval(getSpecifierRange(flag.getPosition(), 1));
1464}
1465
1466void CheckPrintfHandler::HandleIgnoredFlag(
1467                                const analyze_printf::PrintfSpecifier &FS,
1468                                const analyze_printf::OptionalFlag &ignoredFlag,
1469                                const analyze_printf::OptionalFlag &flag,
1470                                const char *startSpecifier,
1471                                unsigned specifierLen) {
1472  // Warn about ignored flag with a fixit removal.
1473  S.Diag(getLocationOfByte(ignoredFlag.getPosition()),
1474      diag::warn_printf_ignored_flag)
1475    << ignoredFlag.toString() << flag.toString()
1476    << getSpecifierRange(startSpecifier, specifierLen)
1477    << FixItHint::CreateRemoval(getSpecifierRange(
1478        ignoredFlag.getPosition(), 1));
1479}
1480
1481bool
1482CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
1483                                            &FS,
1484                                          const char *startSpecifier,
1485                                          unsigned specifierLen) {
1486
1487  using namespace analyze_format_string;
1488  using namespace analyze_printf;
1489  const PrintfConversionSpecifier &CS = FS.getConversionSpecifier();
1490
1491  if (FS.consumesDataArgument()) {
1492    if (atFirstArg) {
1493        atFirstArg = false;
1494        usesPositionalArgs = FS.usesPositionalArg();
1495    }
1496    else if (usesPositionalArgs != FS.usesPositionalArg()) {
1497      // Cannot mix-and-match positional and non-positional arguments.
1498      S.Diag(getLocationOfByte(CS.getStart()),
1499             diag::warn_format_mix_positional_nonpositional_args)
1500        << getSpecifierRange(startSpecifier, specifierLen);
1501      return false;
1502    }
1503  }
1504
1505  // First check if the field width, precision, and conversion specifier
1506  // have matching data arguments.
1507  if (!HandleAmount(FS.getFieldWidth(), /* field width */ 0,
1508                    startSpecifier, specifierLen)) {
1509    return false;
1510  }
1511
1512  if (!HandleAmount(FS.getPrecision(), /* precision */ 1,
1513                    startSpecifier, specifierLen)) {
1514    return false;
1515  }
1516
1517  if (!CS.consumesDataArgument()) {
1518    // FIXME: Technically specifying a precision or field width here
1519    // makes no sense.  Worth issuing a warning at some point.
1520    return true;
1521  }
1522
1523  // Consume the argument.
1524  unsigned argIndex = FS.getArgIndex();
1525  if (argIndex < NumDataArgs) {
1526    // The check to see if the argIndex is valid will come later.
1527    // We set the bit here because we may exit early from this
1528    // function if we encounter some other error.
1529    CoveredArgs.set(argIndex);
1530  }
1531
1532  // Check for using an Objective-C specific conversion specifier
1533  // in a non-ObjC literal.
1534  if (!IsObjCLiteral && CS.isObjCArg()) {
1535    return HandleInvalidPrintfConversionSpecifier(FS, startSpecifier,
1536                                                  specifierLen);
1537  }
1538
1539  // Check for invalid use of field width
1540  if (!FS.hasValidFieldWidth()) {
1541    HandleInvalidAmount(FS, FS.getFieldWidth(), /* field width */ 0,
1542        startSpecifier, specifierLen);
1543  }
1544
1545  // Check for invalid use of precision
1546  if (!FS.hasValidPrecision()) {
1547    HandleInvalidAmount(FS, FS.getPrecision(), /* precision */ 1,
1548        startSpecifier, specifierLen);
1549  }
1550
1551  // Check each flag does not conflict with any other component.
1552  if (!FS.hasValidThousandsGroupingPrefix())
1553    HandleFlag(FS, FS.hasThousandsGrouping(), startSpecifier, specifierLen);
1554  if (!FS.hasValidLeadingZeros())
1555    HandleFlag(FS, FS.hasLeadingZeros(), startSpecifier, specifierLen);
1556  if (!FS.hasValidPlusPrefix())
1557    HandleFlag(FS, FS.hasPlusPrefix(), startSpecifier, specifierLen);
1558  if (!FS.hasValidSpacePrefix())
1559    HandleFlag(FS, FS.hasSpacePrefix(), startSpecifier, specifierLen);
1560  if (!FS.hasValidAlternativeForm())
1561    HandleFlag(FS, FS.hasAlternativeForm(), startSpecifier, specifierLen);
1562  if (!FS.hasValidLeftJustified())
1563    HandleFlag(FS, FS.isLeftJustified(), startSpecifier, specifierLen);
1564
1565  // Check that flags are not ignored by another flag
1566  if (FS.hasSpacePrefix() && FS.hasPlusPrefix()) // ' ' ignored by '+'
1567    HandleIgnoredFlag(FS, FS.hasSpacePrefix(), FS.hasPlusPrefix(),
1568        startSpecifier, specifierLen);
1569  if (FS.hasLeadingZeros() && FS.isLeftJustified()) // '0' ignored by '-'
1570    HandleIgnoredFlag(FS, FS.hasLeadingZeros(), FS.isLeftJustified(),
1571            startSpecifier, specifierLen);
1572
1573  // Check the length modifier is valid with the given conversion specifier.
1574  const LengthModifier &LM = FS.getLengthModifier();
1575  if (!FS.hasValidLengthModifier())
1576    S.Diag(getLocationOfByte(LM.getStart()),
1577        diag::warn_format_nonsensical_length)
1578      << LM.toString() << CS.toString()
1579      << getSpecifierRange(startSpecifier, specifierLen)
1580      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1581          LM.getLength()));
1582
1583  // Are we using '%n'?
1584  if (CS.getKind() == ConversionSpecifier::nArg) {
1585    // Issue a warning about this being a possible security issue.
1586    S.Diag(getLocationOfByte(CS.getStart()), diag::warn_printf_write_back)
1587      << getSpecifierRange(startSpecifier, specifierLen);
1588    // Continue checking the other format specifiers.
1589    return true;
1590  }
1591
1592  // The remaining checks depend on the data arguments.
1593  if (HasVAListArg)
1594    return true;
1595
1596  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
1597    return false;
1598
1599  // Now type check the data expression that matches the
1600  // format specifier.
1601  const Expr *Ex = getDataArg(argIndex);
1602  const analyze_printf::ArgTypeResult &ATR = FS.getArgType(S.Context);
1603  if (ATR.isValid() && !ATR.matchesType(S.Context, Ex->getType())) {
1604    // Check if we didn't match because of an implicit cast from a 'char'
1605    // or 'short' to an 'int'.  This is done because printf is a varargs
1606    // function.
1607    if (const ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(Ex))
1608      if (ICE->getType() == S.Context.IntTy) {
1609        // All further checking is done on the subexpression.
1610        Ex = ICE->getSubExpr();
1611        if (ATR.matchesType(S.Context, Ex->getType()))
1612          return true;
1613      }
1614
1615    // We may be able to offer a FixItHint if it is a supported type.
1616    PrintfSpecifier fixedFS = FS;
1617    bool success = fixedFS.fixType(Ex->getType());
1618
1619    if (success) {
1620      // Get the fix string from the fixed format specifier
1621      llvm::SmallString<128> buf;
1622      llvm::raw_svector_ostream os(buf);
1623      fixedFS.toString(os);
1624
1625      // FIXME: getRepresentativeType() perhaps should return a string
1626      // instead of a QualType to better handle when the representative
1627      // type is 'wint_t' (which is defined in the system headers).
1628      S.Diag(getLocationOfByte(CS.getStart()),
1629          diag::warn_printf_conversion_argument_type_mismatch)
1630        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1631        << getSpecifierRange(startSpecifier, specifierLen)
1632        << Ex->getSourceRange()
1633        << FixItHint::CreateReplacement(
1634            getSpecifierRange(startSpecifier, specifierLen),
1635            os.str());
1636    }
1637    else {
1638      S.Diag(getLocationOfByte(CS.getStart()),
1639             diag::warn_printf_conversion_argument_type_mismatch)
1640        << ATR.getRepresentativeType(S.Context) << Ex->getType()
1641        << getSpecifierRange(startSpecifier, specifierLen)
1642        << Ex->getSourceRange();
1643    }
1644  }
1645
1646  return true;
1647}
1648
1649//===--- CHECK: Scanf format string checking ------------------------------===//
1650
1651namespace {
1652class CheckScanfHandler : public CheckFormatHandler {
1653public:
1654  CheckScanfHandler(Sema &s, const StringLiteral *fexpr,
1655                    const Expr *origFormatExpr, unsigned firstDataArg,
1656                    unsigned numDataArgs, bool isObjCLiteral,
1657                    const char *beg, bool hasVAListArg,
1658                    const CallExpr *theCall, unsigned formatIdx)
1659  : CheckFormatHandler(s, fexpr, origFormatExpr, firstDataArg,
1660                       numDataArgs, isObjCLiteral, beg, hasVAListArg,
1661                       theCall, formatIdx) {}
1662
1663  bool HandleScanfSpecifier(const analyze_scanf::ScanfSpecifier &FS,
1664                            const char *startSpecifier,
1665                            unsigned specifierLen);
1666
1667  bool HandleInvalidScanfConversionSpecifier(
1668          const analyze_scanf::ScanfSpecifier &FS,
1669          const char *startSpecifier,
1670          unsigned specifierLen);
1671
1672  void HandleIncompleteScanList(const char *start, const char *end);
1673};
1674}
1675
1676void CheckScanfHandler::HandleIncompleteScanList(const char *start,
1677                                                 const char *end) {
1678  S.Diag(getLocationOfByte(end), diag::warn_scanf_scanlist_incomplete)
1679    << getSpecifierRange(start, end - start);
1680}
1681
1682bool CheckScanfHandler::HandleInvalidScanfConversionSpecifier(
1683                                        const analyze_scanf::ScanfSpecifier &FS,
1684                                        const char *startSpecifier,
1685                                        unsigned specifierLen) {
1686
1687  const analyze_scanf::ScanfConversionSpecifier &CS =
1688    FS.getConversionSpecifier();
1689
1690  return HandleInvalidConversionSpecifier(FS.getArgIndex(),
1691                                          getLocationOfByte(CS.getStart()),
1692                                          startSpecifier, specifierLen,
1693                                          CS.getStart(), CS.getLength());
1694}
1695
1696bool CheckScanfHandler::HandleScanfSpecifier(
1697                                       const analyze_scanf::ScanfSpecifier &FS,
1698                                       const char *startSpecifier,
1699                                       unsigned specifierLen) {
1700
1701  using namespace analyze_scanf;
1702  using namespace analyze_format_string;
1703
1704  const ScanfConversionSpecifier &CS = FS.getConversionSpecifier();
1705
1706  // Handle case where '%' and '*' don't consume an argument.  These shouldn't
1707  // be used to decide if we are using positional arguments consistently.
1708  if (FS.consumesDataArgument()) {
1709    if (atFirstArg) {
1710      atFirstArg = false;
1711      usesPositionalArgs = FS.usesPositionalArg();
1712    }
1713    else if (usesPositionalArgs != FS.usesPositionalArg()) {
1714      // Cannot mix-and-match positional and non-positional arguments.
1715      S.Diag(getLocationOfByte(CS.getStart()),
1716             diag::warn_format_mix_positional_nonpositional_args)
1717        << getSpecifierRange(startSpecifier, specifierLen);
1718      return false;
1719    }
1720  }
1721
1722  // Check if the field with is non-zero.
1723  const OptionalAmount &Amt = FS.getFieldWidth();
1724  if (Amt.getHowSpecified() == OptionalAmount::Constant) {
1725    if (Amt.getConstantAmount() == 0) {
1726      const CharSourceRange &R = getSpecifierRange(Amt.getStart(),
1727                                                   Amt.getConstantLength());
1728      S.Diag(getLocationOfByte(Amt.getStart()),
1729             diag::warn_scanf_nonzero_width)
1730        << R << FixItHint::CreateRemoval(R);
1731    }
1732  }
1733
1734  if (!FS.consumesDataArgument()) {
1735    // FIXME: Technically specifying a precision or field width here
1736    // makes no sense.  Worth issuing a warning at some point.
1737    return true;
1738  }
1739
1740  // Consume the argument.
1741  unsigned argIndex = FS.getArgIndex();
1742  if (argIndex < NumDataArgs) {
1743      // The check to see if the argIndex is valid will come later.
1744      // We set the bit here because we may exit early from this
1745      // function if we encounter some other error.
1746    CoveredArgs.set(argIndex);
1747  }
1748
1749  // Check the length modifier is valid with the given conversion specifier.
1750  const LengthModifier &LM = FS.getLengthModifier();
1751  if (!FS.hasValidLengthModifier()) {
1752    S.Diag(getLocationOfByte(LM.getStart()),
1753           diag::warn_format_nonsensical_length)
1754      << LM.toString() << CS.toString()
1755      << getSpecifierRange(startSpecifier, specifierLen)
1756      << FixItHint::CreateRemoval(getSpecifierRange(LM.getStart(),
1757                                                    LM.getLength()));
1758  }
1759
1760  // The remaining checks depend on the data arguments.
1761  if (HasVAListArg)
1762    return true;
1763
1764  if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
1765    return false;
1766
1767  // FIXME: Check that the argument type matches the format specifier.
1768
1769  return true;
1770}
1771
1772void Sema::CheckFormatString(const StringLiteral *FExpr,
1773                             const Expr *OrigFormatExpr,
1774                             const CallExpr *TheCall, bool HasVAListArg,
1775                             unsigned format_idx, unsigned firstDataArg,
1776                             bool isPrintf) {
1777
1778  // CHECK: is the format string a wide literal?
1779  if (FExpr->isWide()) {
1780    Diag(FExpr->getLocStart(),
1781         diag::warn_format_string_is_wide_literal)
1782    << OrigFormatExpr->getSourceRange();
1783    return;
1784  }
1785
1786  // Str - The format string.  NOTE: this is NOT null-terminated!
1787  llvm::StringRef StrRef = FExpr->getString();
1788  const char *Str = StrRef.data();
1789  unsigned StrLen = StrRef.size();
1790
1791  // CHECK: empty format string?
1792  if (StrLen == 0) {
1793    Diag(FExpr->getLocStart(), diag::warn_empty_format_string)
1794    << OrigFormatExpr->getSourceRange();
1795    return;
1796  }
1797
1798  if (isPrintf) {
1799    CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1800                         TheCall->getNumArgs() - firstDataArg,
1801                         isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1802                         HasVAListArg, TheCall, format_idx);
1803
1804    if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen))
1805      H.DoneProcessing();
1806  }
1807  else {
1808    CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
1809                        TheCall->getNumArgs() - firstDataArg,
1810                        isa<ObjCStringLiteral>(OrigFormatExpr), Str,
1811                        HasVAListArg, TheCall, format_idx);
1812
1813    if (!analyze_format_string::ParseScanfString(H, Str, Str + StrLen))
1814      H.DoneProcessing();
1815  }
1816}
1817
1818//===--- CHECK: Standard memory functions ---------------------------------===//
1819
1820/// \brief Determine whether the given type is a dynamic class type (e.g.,
1821/// whether it has a vtable).
1822static bool isDynamicClassType(QualType T) {
1823  if (CXXRecordDecl *Record = T->getAsCXXRecordDecl())
1824    if (CXXRecordDecl *Definition = Record->getDefinition())
1825      if (Definition->isDynamicClass())
1826        return true;
1827
1828  return false;
1829}
1830
1831/// \brief If E is a sizeof expression returns the argument expression,
1832/// otherwise returns NULL.
1833static const Expr *getSizeOfExprArg(const Expr* E) {
1834  if (const UnaryExprOrTypeTraitExpr *SizeOf =
1835      dyn_cast<UnaryExprOrTypeTraitExpr>(E))
1836    if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType())
1837      return SizeOf->getArgumentExpr()->IgnoreParenImpCasts();
1838
1839  return 0;
1840}
1841
1842/// \brief If E is a sizeof expression returns the argument type.
1843static QualType getSizeOfArgType(const Expr* E) {
1844  if (const UnaryExprOrTypeTraitExpr *SizeOf =
1845      dyn_cast<UnaryExprOrTypeTraitExpr>(E))
1846    if (SizeOf->getKind() == clang::UETT_SizeOf)
1847      return SizeOf->getTypeOfArgument();
1848
1849  return QualType();
1850}
1851
1852/// \brief Check for dangerous or invalid arguments to memset().
1853///
1854/// This issues warnings on known problematic, dangerous or unspecified
1855/// arguments to the standard 'memset', 'memcpy', and 'memmove' function calls.
1856///
1857/// \param Call The call expression to diagnose.
1858void Sema::CheckMemsetcpymoveArguments(const CallExpr *Call,
1859                                       const IdentifierInfo *FnName) {
1860  // It is possible to have a non-standard definition of memset.  Validate
1861  // we have the proper number of arguments, and if not, abort further
1862  // checking.
1863  if (Call->getNumArgs() != 3)
1864    return;
1865
1866  unsigned LastArg = FnName->isStr("memset")? 1 : 2;
1867  const Expr *LenExpr = Call->getArg(2)->IgnoreParenImpCasts();
1868
1869  // We have special checking when the length is a sizeof expression.
1870  QualType SizeOfArgTy = getSizeOfArgType(LenExpr);
1871  const Expr *SizeOfArg = getSizeOfExprArg(LenExpr);
1872  llvm::FoldingSetNodeID SizeOfArgID;
1873
1874  for (unsigned ArgIdx = 0; ArgIdx != LastArg; ++ArgIdx) {
1875    const Expr *Dest = Call->getArg(ArgIdx)->IgnoreParenImpCasts();
1876    SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange();
1877
1878    QualType DestTy = Dest->getType();
1879    if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) {
1880      QualType PointeeTy = DestPtrTy->getPointeeType();
1881
1882      // Never warn about void type pointers. This can be used to suppress
1883      // false positives.
1884      if (PointeeTy->isVoidType())
1885        continue;
1886
1887      // Catch "memset(p, 0, sizeof(p))" -- needs to be sizeof(*p). Do this by
1888      // actually comparing the expressions for equality. Because computing the
1889      // expression IDs can be expensive, we only do this if the diagnostic is
1890      // enabled.
1891      if (SizeOfArg &&
1892          Diags.getDiagnosticLevel(diag::warn_sizeof_pointer_expr_memaccess,
1893                                   SizeOfArg->getExprLoc())) {
1894        // We only compute IDs for expressions if the warning is enabled, and
1895        // cache the sizeof arg's ID.
1896        if (SizeOfArgID == llvm::FoldingSetNodeID())
1897          SizeOfArg->Profile(SizeOfArgID, Context, true);
1898        llvm::FoldingSetNodeID DestID;
1899        Dest->Profile(DestID, Context, true);
1900        if (DestID == SizeOfArgID) {
1901          unsigned ActionIdx = 0; // Default is to suggest dereferencing.
1902          if (const UnaryOperator *UnaryOp = dyn_cast<UnaryOperator>(Dest))
1903            if (UnaryOp->getOpcode() == UO_AddrOf)
1904              ActionIdx = 1; // If its an address-of operator, just remove it.
1905          if (Context.getTypeSize(PointeeTy) == Context.getCharWidth())
1906            ActionIdx = 2; // If the pointee's size is sizeof(char),
1907                           // suggest an explicit length.
1908          DiagRuntimeBehavior(SizeOfArg->getExprLoc(), Dest,
1909                              PDiag(diag::warn_sizeof_pointer_expr_memaccess)
1910                                << FnName << ArgIdx << ActionIdx
1911                                << Dest->getSourceRange()
1912                                << SizeOfArg->getSourceRange());
1913          break;
1914        }
1915      }
1916
1917      // Also check for cases where the sizeof argument is the exact same
1918      // type as the memory argument, and where it points to a user-defined
1919      // record type.
1920      if (SizeOfArgTy != QualType()) {
1921        if (PointeeTy->isRecordType() &&
1922            Context.typesAreCompatible(SizeOfArgTy, DestTy)) {
1923          DiagRuntimeBehavior(LenExpr->getExprLoc(), Dest,
1924                              PDiag(diag::warn_sizeof_pointer_type_memaccess)
1925                                << FnName << SizeOfArgTy << ArgIdx
1926                                << PointeeTy << Dest->getSourceRange()
1927                                << LenExpr->getSourceRange());
1928          break;
1929        }
1930      }
1931
1932      unsigned DiagID;
1933
1934      // Always complain about dynamic classes.
1935      if (isDynamicClassType(PointeeTy))
1936        DiagID = diag::warn_dyn_class_memaccess;
1937      else if (PointeeTy.hasNonTrivialObjCLifetime() &&
1938               !FnName->isStr("memset"))
1939        DiagID = diag::warn_arc_object_memaccess;
1940      else
1941        continue;
1942
1943      DiagRuntimeBehavior(
1944        Dest->getExprLoc(), Dest,
1945        PDiag(DiagID)
1946          << ArgIdx << FnName << PointeeTy
1947          << Call->getCallee()->getSourceRange());
1948
1949      DiagRuntimeBehavior(
1950        Dest->getExprLoc(), Dest,
1951        PDiag(diag::note_bad_memaccess_silence)
1952          << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)"));
1953      break;
1954    }
1955  }
1956}
1957
1958//===--- CHECK: Return Address of Stack Variable --------------------------===//
1959
1960static Expr *EvalVal(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars);
1961static Expr *EvalAddr(Expr* E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars);
1962
1963/// CheckReturnStackAddr - Check if a return statement returns the address
1964///   of a stack variable.
1965void
1966Sema::CheckReturnStackAddr(Expr *RetValExp, QualType lhsType,
1967                           SourceLocation ReturnLoc) {
1968
1969  Expr *stackE = 0;
1970  llvm::SmallVector<DeclRefExpr *, 8> refVars;
1971
1972  // Perform checking for returned stack addresses, local blocks,
1973  // label addresses or references to temporaries.
1974  if (lhsType->isPointerType() ||
1975      (!getLangOptions().ObjCAutoRefCount && lhsType->isBlockPointerType())) {
1976    stackE = EvalAddr(RetValExp, refVars);
1977  } else if (lhsType->isReferenceType()) {
1978    stackE = EvalVal(RetValExp, refVars);
1979  }
1980
1981  if (stackE == 0)
1982    return; // Nothing suspicious was found.
1983
1984  SourceLocation diagLoc;
1985  SourceRange diagRange;
1986  if (refVars.empty()) {
1987    diagLoc = stackE->getLocStart();
1988    diagRange = stackE->getSourceRange();
1989  } else {
1990    // We followed through a reference variable. 'stackE' contains the
1991    // problematic expression but we will warn at the return statement pointing
1992    // at the reference variable. We will later display the "trail" of
1993    // reference variables using notes.
1994    diagLoc = refVars[0]->getLocStart();
1995    diagRange = refVars[0]->getSourceRange();
1996  }
1997
1998  if (DeclRefExpr *DR = dyn_cast<DeclRefExpr>(stackE)) { //address of local var.
1999    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_stack_ref
2000                                             : diag::warn_ret_stack_addr)
2001     << DR->getDecl()->getDeclName() << diagRange;
2002  } else if (isa<BlockExpr>(stackE)) { // local block.
2003    Diag(diagLoc, diag::err_ret_local_block) << diagRange;
2004  } else if (isa<AddrLabelExpr>(stackE)) { // address of label.
2005    Diag(diagLoc, diag::warn_ret_addr_label) << diagRange;
2006  } else { // local temporary.
2007    Diag(diagLoc, lhsType->isReferenceType() ? diag::warn_ret_local_temp_ref
2008                                             : diag::warn_ret_local_temp_addr)
2009     << diagRange;
2010  }
2011
2012  // Display the "trail" of reference variables that we followed until we
2013  // found the problematic expression using notes.
2014  for (unsigned i = 0, e = refVars.size(); i != e; ++i) {
2015    VarDecl *VD = cast<VarDecl>(refVars[i]->getDecl());
2016    // If this var binds to another reference var, show the range of the next
2017    // var, otherwise the var binds to the problematic expression, in which case
2018    // show the range of the expression.
2019    SourceRange range = (i < e-1) ? refVars[i+1]->getSourceRange()
2020                                  : stackE->getSourceRange();
2021    Diag(VD->getLocation(), diag::note_ref_var_local_bind)
2022      << VD->getDeclName() << range;
2023  }
2024}
2025
2026/// EvalAddr - EvalAddr and EvalVal are mutually recursive functions that
2027///  check if the expression in a return statement evaluates to an address
2028///  to a location on the stack, a local block, an address of a label, or a
2029///  reference to local temporary. The recursion is used to traverse the
2030///  AST of the return expression, with recursion backtracking when we
2031///  encounter a subexpression that (1) clearly does not lead to one of the
2032///  above problematic expressions (2) is something we cannot determine leads to
2033///  a problematic expression based on such local checking.
2034///
2035///  Both EvalAddr and EvalVal follow through reference variables to evaluate
2036///  the expression that they point to. Such variables are added to the
2037///  'refVars' vector so that we know what the reference variable "trail" was.
2038///
2039///  EvalAddr processes expressions that are pointers that are used as
2040///  references (and not L-values).  EvalVal handles all other values.
2041///  At the base case of the recursion is a check for the above problematic
2042///  expressions.
2043///
2044///  This implementation handles:
2045///
2046///   * pointer-to-pointer casts
2047///   * implicit conversions from array references to pointers
2048///   * taking the address of fields
2049///   * arbitrary interplay between "&" and "*" operators
2050///   * pointer arithmetic from an address of a stack variable
2051///   * taking the address of an array element where the array is on the stack
2052static Expr *EvalAddr(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars) {
2053  if (E->isTypeDependent())
2054      return NULL;
2055
2056  // We should only be called for evaluating pointer expressions.
2057  assert((E->getType()->isAnyPointerType() ||
2058          E->getType()->isBlockPointerType() ||
2059          E->getType()->isObjCQualifiedIdType()) &&
2060         "EvalAddr only works on pointers");
2061
2062  E = E->IgnoreParens();
2063
2064  // Our "symbolic interpreter" is just a dispatch off the currently
2065  // viewed AST node.  We then recursively traverse the AST by calling
2066  // EvalAddr and EvalVal appropriately.
2067  switch (E->getStmtClass()) {
2068  case Stmt::DeclRefExprClass: {
2069    DeclRefExpr *DR = cast<DeclRefExpr>(E);
2070
2071    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
2072      // If this is a reference variable, follow through to the expression that
2073      // it points to.
2074      if (V->hasLocalStorage() &&
2075          V->getType()->isReferenceType() && V->hasInit()) {
2076        // Add the reference variable to the "trail".
2077        refVars.push_back(DR);
2078        return EvalAddr(V->getInit(), refVars);
2079      }
2080
2081    return NULL;
2082  }
2083
2084  case Stmt::UnaryOperatorClass: {
2085    // The only unary operator that make sense to handle here
2086    // is AddrOf.  All others don't make sense as pointers.
2087    UnaryOperator *U = cast<UnaryOperator>(E);
2088
2089    if (U->getOpcode() == UO_AddrOf)
2090      return EvalVal(U->getSubExpr(), refVars);
2091    else
2092      return NULL;
2093  }
2094
2095  case Stmt::BinaryOperatorClass: {
2096    // Handle pointer arithmetic.  All other binary operators are not valid
2097    // in this context.
2098    BinaryOperator *B = cast<BinaryOperator>(E);
2099    BinaryOperatorKind op = B->getOpcode();
2100
2101    if (op != BO_Add && op != BO_Sub)
2102      return NULL;
2103
2104    Expr *Base = B->getLHS();
2105
2106    // Determine which argument is the real pointer base.  It could be
2107    // the RHS argument instead of the LHS.
2108    if (!Base->getType()->isPointerType()) Base = B->getRHS();
2109
2110    assert (Base->getType()->isPointerType());
2111    return EvalAddr(Base, refVars);
2112  }
2113
2114  // For conditional operators we need to see if either the LHS or RHS are
2115  // valid DeclRefExpr*s.  If one of them is valid, we return it.
2116  case Stmt::ConditionalOperatorClass: {
2117    ConditionalOperator *C = cast<ConditionalOperator>(E);
2118
2119    // Handle the GNU extension for missing LHS.
2120    if (Expr *lhsExpr = C->getLHS()) {
2121    // In C++, we can have a throw-expression, which has 'void' type.
2122      if (!lhsExpr->getType()->isVoidType())
2123        if (Expr* LHS = EvalAddr(lhsExpr, refVars))
2124          return LHS;
2125    }
2126
2127    // In C++, we can have a throw-expression, which has 'void' type.
2128    if (C->getRHS()->getType()->isVoidType())
2129      return NULL;
2130
2131    return EvalAddr(C->getRHS(), refVars);
2132  }
2133
2134  case Stmt::BlockExprClass:
2135    if (cast<BlockExpr>(E)->getBlockDecl()->hasCaptures())
2136      return E; // local block.
2137    return NULL;
2138
2139  case Stmt::AddrLabelExprClass:
2140    return E; // address of label.
2141
2142  // For casts, we need to handle conversions from arrays to
2143  // pointer values, and pointer-to-pointer conversions.
2144  case Stmt::ImplicitCastExprClass:
2145  case Stmt::CStyleCastExprClass:
2146  case Stmt::CXXFunctionalCastExprClass:
2147  case Stmt::ObjCBridgedCastExprClass: {
2148    Expr* SubExpr = cast<CastExpr>(E)->getSubExpr();
2149    QualType T = SubExpr->getType();
2150
2151    if (SubExpr->getType()->isPointerType() ||
2152        SubExpr->getType()->isBlockPointerType() ||
2153        SubExpr->getType()->isObjCQualifiedIdType())
2154      return EvalAddr(SubExpr, refVars);
2155    else if (T->isArrayType())
2156      return EvalVal(SubExpr, refVars);
2157    else
2158      return 0;
2159  }
2160
2161  // C++ casts.  For dynamic casts, static casts, and const casts, we
2162  // are always converting from a pointer-to-pointer, so we just blow
2163  // through the cast.  In the case the dynamic cast doesn't fail (and
2164  // return NULL), we take the conservative route and report cases
2165  // where we return the address of a stack variable.  For Reinterpre
2166  // FIXME: The comment about is wrong; we're not always converting
2167  // from pointer to pointer. I'm guessing that this code should also
2168  // handle references to objects.
2169  case Stmt::CXXStaticCastExprClass:
2170  case Stmt::CXXDynamicCastExprClass:
2171  case Stmt::CXXConstCastExprClass:
2172  case Stmt::CXXReinterpretCastExprClass: {
2173      Expr *S = cast<CXXNamedCastExpr>(E)->getSubExpr();
2174      if (S->getType()->isPointerType() || S->getType()->isBlockPointerType())
2175        return EvalAddr(S, refVars);
2176      else
2177        return NULL;
2178  }
2179
2180  // Everything else: we simply don't reason about them.
2181  default:
2182    return NULL;
2183  }
2184}
2185
2186
2187///  EvalVal - This function is complements EvalAddr in the mutual recursion.
2188///   See the comments for EvalAddr for more details.
2189static Expr *EvalVal(Expr *E, llvm::SmallVectorImpl<DeclRefExpr *> &refVars) {
2190do {
2191  // We should only be called for evaluating non-pointer expressions, or
2192  // expressions with a pointer type that are not used as references but instead
2193  // are l-values (e.g., DeclRefExpr with a pointer type).
2194
2195  // Our "symbolic interpreter" is just a dispatch off the currently
2196  // viewed AST node.  We then recursively traverse the AST by calling
2197  // EvalAddr and EvalVal appropriately.
2198
2199  E = E->IgnoreParens();
2200  switch (E->getStmtClass()) {
2201  case Stmt::ImplicitCastExprClass: {
2202    ImplicitCastExpr *IE = cast<ImplicitCastExpr>(E);
2203    if (IE->getValueKind() == VK_LValue) {
2204      E = IE->getSubExpr();
2205      continue;
2206    }
2207    return NULL;
2208  }
2209
2210  case Stmt::DeclRefExprClass: {
2211    // When we hit a DeclRefExpr we are looking at code that refers to a
2212    // variable's name. If it's not a reference variable we check if it has
2213    // local storage within the function, and if so, return the expression.
2214    DeclRefExpr *DR = cast<DeclRefExpr>(E);
2215
2216    if (VarDecl *V = dyn_cast<VarDecl>(DR->getDecl()))
2217      if (V->hasLocalStorage()) {
2218        if (!V->getType()->isReferenceType())
2219          return DR;
2220
2221        // Reference variable, follow through to the expression that
2222        // it points to.
2223        if (V->hasInit()) {
2224          // Add the reference variable to the "trail".
2225          refVars.push_back(DR);
2226          return EvalVal(V->getInit(), refVars);
2227        }
2228      }
2229
2230    return NULL;
2231  }
2232
2233  case Stmt::UnaryOperatorClass: {
2234    // The only unary operator that make sense to handle here
2235    // is Deref.  All others don't resolve to a "name."  This includes
2236    // handling all sorts of rvalues passed to a unary operator.
2237    UnaryOperator *U = cast<UnaryOperator>(E);
2238
2239    if (U->getOpcode() == UO_Deref)
2240      return EvalAddr(U->getSubExpr(), refVars);
2241
2242    return NULL;
2243  }
2244
2245  case Stmt::ArraySubscriptExprClass: {
2246    // Array subscripts are potential references to data on the stack.  We
2247    // retrieve the DeclRefExpr* for the array variable if it indeed
2248    // has local storage.
2249    return EvalAddr(cast<ArraySubscriptExpr>(E)->getBase(), refVars);
2250  }
2251
2252  case Stmt::ConditionalOperatorClass: {
2253    // For conditional operators we need to see if either the LHS or RHS are
2254    // non-NULL Expr's.  If one is non-NULL, we return it.
2255    ConditionalOperator *C = cast<ConditionalOperator>(E);
2256
2257    // Handle the GNU extension for missing LHS.
2258    if (Expr *lhsExpr = C->getLHS())
2259      if (Expr *LHS = EvalVal(lhsExpr, refVars))
2260        return LHS;
2261
2262    return EvalVal(C->getRHS(), refVars);
2263  }
2264
2265  // Accesses to members are potential references to data on the stack.
2266  case Stmt::MemberExprClass: {
2267    MemberExpr *M = cast<MemberExpr>(E);
2268
2269    // Check for indirect access.  We only want direct field accesses.
2270    if (M->isArrow())
2271      return NULL;
2272
2273    // Check whether the member type is itself a reference, in which case
2274    // we're not going to refer to the member, but to what the member refers to.
2275    if (M->getMemberDecl()->getType()->isReferenceType())
2276      return NULL;
2277
2278    return EvalVal(M->getBase(), refVars);
2279  }
2280
2281  default:
2282    // Check that we don't return or take the address of a reference to a
2283    // temporary. This is only useful in C++.
2284    if (!E->isTypeDependent() && E->isRValue())
2285      return E;
2286
2287    // Everything else: we simply don't reason about them.
2288    return NULL;
2289  }
2290} while (true);
2291}
2292
2293//===--- CHECK: Floating-Point comparisons (-Wfloat-equal) ---------------===//
2294
2295/// Check for comparisons of floating point operands using != and ==.
2296/// Issue a warning if these are no self-comparisons, as they are not likely
2297/// to do what the programmer intended.
2298void Sema::CheckFloatComparison(SourceLocation loc, Expr* lex, Expr *rex) {
2299  bool EmitWarning = true;
2300
2301  Expr* LeftExprSansParen = lex->IgnoreParenImpCasts();
2302  Expr* RightExprSansParen = rex->IgnoreParenImpCasts();
2303
2304  // Special case: check for x == x (which is OK).
2305  // Do not emit warnings for such cases.
2306  if (DeclRefExpr* DRL = dyn_cast<DeclRefExpr>(LeftExprSansParen))
2307    if (DeclRefExpr* DRR = dyn_cast<DeclRefExpr>(RightExprSansParen))
2308      if (DRL->getDecl() == DRR->getDecl())
2309        EmitWarning = false;
2310
2311
2312  // Special case: check for comparisons against literals that can be exactly
2313  //  represented by APFloat.  In such cases, do not emit a warning.  This
2314  //  is a heuristic: often comparison against such literals are used to
2315  //  detect if a value in a variable has not changed.  This clearly can
2316  //  lead to false negatives.
2317  if (EmitWarning) {
2318    if (FloatingLiteral* FLL = dyn_cast<FloatingLiteral>(LeftExprSansParen)) {
2319      if (FLL->isExact())
2320        EmitWarning = false;
2321    } else
2322      if (FloatingLiteral* FLR = dyn_cast<FloatingLiteral>(RightExprSansParen)){
2323        if (FLR->isExact())
2324          EmitWarning = false;
2325    }
2326  }
2327
2328  // Check for comparisons with builtin types.
2329  if (EmitWarning)
2330    if (CallExpr* CL = dyn_cast<CallExpr>(LeftExprSansParen))
2331      if (CL->isBuiltinCall(Context))
2332        EmitWarning = false;
2333
2334  if (EmitWarning)
2335    if (CallExpr* CR = dyn_cast<CallExpr>(RightExprSansParen))
2336      if (CR->isBuiltinCall(Context))
2337        EmitWarning = false;
2338
2339  // Emit the diagnostic.
2340  if (EmitWarning)
2341    Diag(loc, diag::warn_floatingpoint_eq)
2342      << lex->getSourceRange() << rex->getSourceRange();
2343}
2344
2345//===--- CHECK: Integer mixed-sign comparisons (-Wsign-compare) --------===//
2346//===--- CHECK: Lossy implicit conversions (-Wconversion) --------------===//
2347
2348namespace {
2349
2350/// Structure recording the 'active' range of an integer-valued
2351/// expression.
2352struct IntRange {
2353  /// The number of bits active in the int.
2354  unsigned Width;
2355
2356  /// True if the int is known not to have negative values.
2357  bool NonNegative;
2358
2359  IntRange(unsigned Width, bool NonNegative)
2360    : Width(Width), NonNegative(NonNegative)
2361  {}
2362
2363  /// Returns the range of the bool type.
2364  static IntRange forBoolType() {
2365    return IntRange(1, true);
2366  }
2367
2368  /// Returns the range of an opaque value of the given integral type.
2369  static IntRange forValueOfType(ASTContext &C, QualType T) {
2370    return forValueOfCanonicalType(C,
2371                          T->getCanonicalTypeInternal().getTypePtr());
2372  }
2373
2374  /// Returns the range of an opaque value of a canonical integral type.
2375  static IntRange forValueOfCanonicalType(ASTContext &C, const Type *T) {
2376    assert(T->isCanonicalUnqualified());
2377
2378    if (const VectorType *VT = dyn_cast<VectorType>(T))
2379      T = VT->getElementType().getTypePtr();
2380    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2381      T = CT->getElementType().getTypePtr();
2382
2383    // For enum types, use the known bit width of the enumerators.
2384    if (const EnumType *ET = dyn_cast<EnumType>(T)) {
2385      EnumDecl *Enum = ET->getDecl();
2386      if (!Enum->isDefinition())
2387        return IntRange(C.getIntWidth(QualType(T, 0)), false);
2388
2389      unsigned NumPositive = Enum->getNumPositiveBits();
2390      unsigned NumNegative = Enum->getNumNegativeBits();
2391
2392      return IntRange(std::max(NumPositive, NumNegative), NumNegative == 0);
2393    }
2394
2395    const BuiltinType *BT = cast<BuiltinType>(T);
2396    assert(BT->isInteger());
2397
2398    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2399  }
2400
2401  /// Returns the "target" range of a canonical integral type, i.e.
2402  /// the range of values expressible in the type.
2403  ///
2404  /// This matches forValueOfCanonicalType except that enums have the
2405  /// full range of their type, not the range of their enumerators.
2406  static IntRange forTargetOfCanonicalType(ASTContext &C, const Type *T) {
2407    assert(T->isCanonicalUnqualified());
2408
2409    if (const VectorType *VT = dyn_cast<VectorType>(T))
2410      T = VT->getElementType().getTypePtr();
2411    if (const ComplexType *CT = dyn_cast<ComplexType>(T))
2412      T = CT->getElementType().getTypePtr();
2413    if (const EnumType *ET = dyn_cast<EnumType>(T))
2414      T = ET->getDecl()->getIntegerType().getTypePtr();
2415
2416    const BuiltinType *BT = cast<BuiltinType>(T);
2417    assert(BT->isInteger());
2418
2419    return IntRange(C.getIntWidth(QualType(T, 0)), BT->isUnsignedInteger());
2420  }
2421
2422  /// Returns the supremum of two ranges: i.e. their conservative merge.
2423  static IntRange join(IntRange L, IntRange R) {
2424    return IntRange(std::max(L.Width, R.Width),
2425                    L.NonNegative && R.NonNegative);
2426  }
2427
2428  /// Returns the infinum of two ranges: i.e. their aggressive merge.
2429  static IntRange meet(IntRange L, IntRange R) {
2430    return IntRange(std::min(L.Width, R.Width),
2431                    L.NonNegative || R.NonNegative);
2432  }
2433};
2434
2435IntRange GetValueRange(ASTContext &C, llvm::APSInt &value, unsigned MaxWidth) {
2436  if (value.isSigned() && value.isNegative())
2437    return IntRange(value.getMinSignedBits(), false);
2438
2439  if (value.getBitWidth() > MaxWidth)
2440    value = value.trunc(MaxWidth);
2441
2442  // isNonNegative() just checks the sign bit without considering
2443  // signedness.
2444  return IntRange(value.getActiveBits(), true);
2445}
2446
2447IntRange GetValueRange(ASTContext &C, APValue &result, QualType Ty,
2448                       unsigned MaxWidth) {
2449  if (result.isInt())
2450    return GetValueRange(C, result.getInt(), MaxWidth);
2451
2452  if (result.isVector()) {
2453    IntRange R = GetValueRange(C, result.getVectorElt(0), Ty, MaxWidth);
2454    for (unsigned i = 1, e = result.getVectorLength(); i != e; ++i) {
2455      IntRange El = GetValueRange(C, result.getVectorElt(i), Ty, MaxWidth);
2456      R = IntRange::join(R, El);
2457    }
2458    return R;
2459  }
2460
2461  if (result.isComplexInt()) {
2462    IntRange R = GetValueRange(C, result.getComplexIntReal(), MaxWidth);
2463    IntRange I = GetValueRange(C, result.getComplexIntImag(), MaxWidth);
2464    return IntRange::join(R, I);
2465  }
2466
2467  // This can happen with lossless casts to intptr_t of "based" lvalues.
2468  // Assume it might use arbitrary bits.
2469  // FIXME: The only reason we need to pass the type in here is to get
2470  // the sign right on this one case.  It would be nice if APValue
2471  // preserved this.
2472  assert(result.isLValue());
2473  return IntRange(MaxWidth, Ty->isUnsignedIntegerOrEnumerationType());
2474}
2475
2476/// Pseudo-evaluate the given integer expression, estimating the
2477/// range of values it might take.
2478///
2479/// \param MaxWidth - the width to which the value will be truncated
2480IntRange GetExprRange(ASTContext &C, Expr *E, unsigned MaxWidth) {
2481  E = E->IgnoreParens();
2482
2483  // Try a full evaluation first.
2484  Expr::EvalResult result;
2485  if (E->Evaluate(result, C))
2486    return GetValueRange(C, result.Val, E->getType(), MaxWidth);
2487
2488  // I think we only want to look through implicit casts here; if the
2489  // user has an explicit widening cast, we should treat the value as
2490  // being of the new, wider type.
2491  if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) {
2492    if (CE->getCastKind() == CK_NoOp)
2493      return GetExprRange(C, CE->getSubExpr(), MaxWidth);
2494
2495    IntRange OutputTypeRange = IntRange::forValueOfType(C, CE->getType());
2496
2497    bool isIntegerCast = (CE->getCastKind() == CK_IntegralCast);
2498
2499    // Assume that non-integer casts can span the full range of the type.
2500    if (!isIntegerCast)
2501      return OutputTypeRange;
2502
2503    IntRange SubRange
2504      = GetExprRange(C, CE->getSubExpr(),
2505                     std::min(MaxWidth, OutputTypeRange.Width));
2506
2507    // Bail out if the subexpr's range is as wide as the cast type.
2508    if (SubRange.Width >= OutputTypeRange.Width)
2509      return OutputTypeRange;
2510
2511    // Otherwise, we take the smaller width, and we're non-negative if
2512    // either the output type or the subexpr is.
2513    return IntRange(SubRange.Width,
2514                    SubRange.NonNegative || OutputTypeRange.NonNegative);
2515  }
2516
2517  if (ConditionalOperator *CO = dyn_cast<ConditionalOperator>(E)) {
2518    // If we can fold the condition, just take that operand.
2519    bool CondResult;
2520    if (CO->getCond()->EvaluateAsBooleanCondition(CondResult, C))
2521      return GetExprRange(C, CondResult ? CO->getTrueExpr()
2522                                        : CO->getFalseExpr(),
2523                          MaxWidth);
2524
2525    // Otherwise, conservatively merge.
2526    IntRange L = GetExprRange(C, CO->getTrueExpr(), MaxWidth);
2527    IntRange R = GetExprRange(C, CO->getFalseExpr(), MaxWidth);
2528    return IntRange::join(L, R);
2529  }
2530
2531  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
2532    switch (BO->getOpcode()) {
2533
2534    // Boolean-valued operations are single-bit and positive.
2535    case BO_LAnd:
2536    case BO_LOr:
2537    case BO_LT:
2538    case BO_GT:
2539    case BO_LE:
2540    case BO_GE:
2541    case BO_EQ:
2542    case BO_NE:
2543      return IntRange::forBoolType();
2544
2545    // The type of these compound assignments is the type of the LHS,
2546    // so the RHS is not necessarily an integer.
2547    case BO_MulAssign:
2548    case BO_DivAssign:
2549    case BO_RemAssign:
2550    case BO_AddAssign:
2551    case BO_SubAssign:
2552      return IntRange::forValueOfType(C, E->getType());
2553
2554    // Operations with opaque sources are black-listed.
2555    case BO_PtrMemD:
2556    case BO_PtrMemI:
2557      return IntRange::forValueOfType(C, E->getType());
2558
2559    // Bitwise-and uses the *infinum* of the two source ranges.
2560    case BO_And:
2561    case BO_AndAssign:
2562      return IntRange::meet(GetExprRange(C, BO->getLHS(), MaxWidth),
2563                            GetExprRange(C, BO->getRHS(), MaxWidth));
2564
2565    // Left shift gets black-listed based on a judgement call.
2566    case BO_Shl:
2567      // ...except that we want to treat '1 << (blah)' as logically
2568      // positive.  It's an important idiom.
2569      if (IntegerLiteral *I
2570            = dyn_cast<IntegerLiteral>(BO->getLHS()->IgnoreParenCasts())) {
2571        if (I->getValue() == 1) {
2572          IntRange R = IntRange::forValueOfType(C, E->getType());
2573          return IntRange(R.Width, /*NonNegative*/ true);
2574        }
2575      }
2576      // fallthrough
2577
2578    case BO_ShlAssign:
2579      return IntRange::forValueOfType(C, E->getType());
2580
2581    // Right shift by a constant can narrow its left argument.
2582    case BO_Shr:
2583    case BO_ShrAssign: {
2584      IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2585
2586      // If the shift amount is a positive constant, drop the width by
2587      // that much.
2588      llvm::APSInt shift;
2589      if (BO->getRHS()->isIntegerConstantExpr(shift, C) &&
2590          shift.isNonNegative()) {
2591        unsigned zext = shift.getZExtValue();
2592        if (zext >= L.Width)
2593          L.Width = (L.NonNegative ? 0 : 1);
2594        else
2595          L.Width -= zext;
2596      }
2597
2598      return L;
2599    }
2600
2601    // Comma acts as its right operand.
2602    case BO_Comma:
2603      return GetExprRange(C, BO->getRHS(), MaxWidth);
2604
2605    // Black-list pointer subtractions.
2606    case BO_Sub:
2607      if (BO->getLHS()->getType()->isPointerType())
2608        return IntRange::forValueOfType(C, E->getType());
2609      // fallthrough
2610
2611    default:
2612      break;
2613    }
2614
2615    // Treat every other operator as if it were closed on the
2616    // narrowest type that encompasses both operands.
2617    IntRange L = GetExprRange(C, BO->getLHS(), MaxWidth);
2618    IntRange R = GetExprRange(C, BO->getRHS(), MaxWidth);
2619    return IntRange::join(L, R);
2620  }
2621
2622  if (UnaryOperator *UO = dyn_cast<UnaryOperator>(E)) {
2623    switch (UO->getOpcode()) {
2624    // Boolean-valued operations are white-listed.
2625    case UO_LNot:
2626      return IntRange::forBoolType();
2627
2628    // Operations with opaque sources are black-listed.
2629    case UO_Deref:
2630    case UO_AddrOf: // should be impossible
2631      return IntRange::forValueOfType(C, E->getType());
2632
2633    default:
2634      return GetExprRange(C, UO->getSubExpr(), MaxWidth);
2635    }
2636  }
2637
2638  if (dyn_cast<OffsetOfExpr>(E)) {
2639    IntRange::forValueOfType(C, E->getType());
2640  }
2641
2642  FieldDecl *BitField = E->getBitField();
2643  if (BitField) {
2644    llvm::APSInt BitWidthAP = BitField->getBitWidth()->EvaluateAsInt(C);
2645    unsigned BitWidth = BitWidthAP.getZExtValue();
2646
2647    return IntRange(BitWidth,
2648                    BitField->getType()->isUnsignedIntegerOrEnumerationType());
2649  }
2650
2651  return IntRange::forValueOfType(C, E->getType());
2652}
2653
2654IntRange GetExprRange(ASTContext &C, Expr *E) {
2655  return GetExprRange(C, E, C.getIntWidth(E->getType()));
2656}
2657
2658/// Checks whether the given value, which currently has the given
2659/// source semantics, has the same value when coerced through the
2660/// target semantics.
2661bool IsSameFloatAfterCast(const llvm::APFloat &value,
2662                          const llvm::fltSemantics &Src,
2663                          const llvm::fltSemantics &Tgt) {
2664  llvm::APFloat truncated = value;
2665
2666  bool ignored;
2667  truncated.convert(Src, llvm::APFloat::rmNearestTiesToEven, &ignored);
2668  truncated.convert(Tgt, llvm::APFloat::rmNearestTiesToEven, &ignored);
2669
2670  return truncated.bitwiseIsEqual(value);
2671}
2672
2673/// Checks whether the given value, which currently has the given
2674/// source semantics, has the same value when coerced through the
2675/// target semantics.
2676///
2677/// The value might be a vector of floats (or a complex number).
2678bool IsSameFloatAfterCast(const APValue &value,
2679                          const llvm::fltSemantics &Src,
2680                          const llvm::fltSemantics &Tgt) {
2681  if (value.isFloat())
2682    return IsSameFloatAfterCast(value.getFloat(), Src, Tgt);
2683
2684  if (value.isVector()) {
2685    for (unsigned i = 0, e = value.getVectorLength(); i != e; ++i)
2686      if (!IsSameFloatAfterCast(value.getVectorElt(i), Src, Tgt))
2687        return false;
2688    return true;
2689  }
2690
2691  assert(value.isComplexFloat());
2692  return (IsSameFloatAfterCast(value.getComplexFloatReal(), Src, Tgt) &&
2693          IsSameFloatAfterCast(value.getComplexFloatImag(), Src, Tgt));
2694}
2695
2696void AnalyzeImplicitConversions(Sema &S, Expr *E, SourceLocation CC);
2697
2698static bool IsZero(Sema &S, Expr *E) {
2699  // Suppress cases where we are comparing against an enum constant.
2700  if (const DeclRefExpr *DR =
2701      dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts()))
2702    if (isa<EnumConstantDecl>(DR->getDecl()))
2703      return false;
2704
2705  // Suppress cases where the '0' value is expanded from a macro.
2706  if (E->getLocStart().isMacroID())
2707    return false;
2708
2709  llvm::APSInt Value;
2710  return E->isIntegerConstantExpr(Value, S.Context) && Value == 0;
2711}
2712
2713static bool HasEnumType(Expr *E) {
2714  // Strip off implicit integral promotions.
2715  while (ImplicitCastExpr *ICE = dyn_cast<ImplicitCastExpr>(E)) {
2716    if (ICE->getCastKind() != CK_IntegralCast &&
2717        ICE->getCastKind() != CK_NoOp)
2718      break;
2719    E = ICE->getSubExpr();
2720  }
2721
2722  return E->getType()->isEnumeralType();
2723}
2724
2725void CheckTrivialUnsignedComparison(Sema &S, BinaryOperator *E) {
2726  BinaryOperatorKind op = E->getOpcode();
2727  if (E->isValueDependent())
2728    return;
2729
2730  if (op == BO_LT && IsZero(S, E->getRHS())) {
2731    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2732      << "< 0" << "false" << HasEnumType(E->getLHS())
2733      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2734  } else if (op == BO_GE && IsZero(S, E->getRHS())) {
2735    S.Diag(E->getOperatorLoc(), diag::warn_lunsigned_always_true_comparison)
2736      << ">= 0" << "true" << HasEnumType(E->getLHS())
2737      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2738  } else if (op == BO_GT && IsZero(S, E->getLHS())) {
2739    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
2740      << "0 >" << "false" << HasEnumType(E->getRHS())
2741      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2742  } else if (op == BO_LE && IsZero(S, E->getLHS())) {
2743    S.Diag(E->getOperatorLoc(), diag::warn_runsigned_always_true_comparison)
2744      << "0 <=" << "true" << HasEnumType(E->getRHS())
2745      << E->getLHS()->getSourceRange() << E->getRHS()->getSourceRange();
2746  }
2747}
2748
2749/// Analyze the operands of the given comparison.  Implements the
2750/// fallback case from AnalyzeComparison.
2751void AnalyzeImpConvsInComparison(Sema &S, BinaryOperator *E) {
2752  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
2753  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
2754}
2755
2756/// \brief Implements -Wsign-compare.
2757///
2758/// \param lex the left-hand expression
2759/// \param rex the right-hand expression
2760/// \param OpLoc the location of the joining operator
2761/// \param BinOpc binary opcode or 0
2762void AnalyzeComparison(Sema &S, BinaryOperator *E) {
2763  // The type the comparison is being performed in.
2764  QualType T = E->getLHS()->getType();
2765  assert(S.Context.hasSameUnqualifiedType(T, E->getRHS()->getType())
2766         && "comparison with mismatched types");
2767
2768  // We don't do anything special if this isn't an unsigned integral
2769  // comparison:  we're only interested in integral comparisons, and
2770  // signed comparisons only happen in cases we don't care to warn about.
2771  //
2772  // We also don't care about value-dependent expressions or expressions
2773  // whose result is a constant.
2774  if (!T->hasUnsignedIntegerRepresentation()
2775      || E->isValueDependent() || E->isIntegerConstantExpr(S.Context))
2776    return AnalyzeImpConvsInComparison(S, E);
2777
2778  Expr *lex = E->getLHS()->IgnoreParenImpCasts();
2779  Expr *rex = E->getRHS()->IgnoreParenImpCasts();
2780
2781  // Check to see if one of the (unmodified) operands is of different
2782  // signedness.
2783  Expr *signedOperand, *unsignedOperand;
2784  if (lex->getType()->hasSignedIntegerRepresentation()) {
2785    assert(!rex->getType()->hasSignedIntegerRepresentation() &&
2786           "unsigned comparison between two signed integer expressions?");
2787    signedOperand = lex;
2788    unsignedOperand = rex;
2789  } else if (rex->getType()->hasSignedIntegerRepresentation()) {
2790    signedOperand = rex;
2791    unsignedOperand = lex;
2792  } else {
2793    CheckTrivialUnsignedComparison(S, E);
2794    return AnalyzeImpConvsInComparison(S, E);
2795  }
2796
2797  // Otherwise, calculate the effective range of the signed operand.
2798  IntRange signedRange = GetExprRange(S.Context, signedOperand);
2799
2800  // Go ahead and analyze implicit conversions in the operands.  Note
2801  // that we skip the implicit conversions on both sides.
2802  AnalyzeImplicitConversions(S, lex, E->getOperatorLoc());
2803  AnalyzeImplicitConversions(S, rex, E->getOperatorLoc());
2804
2805  // If the signed range is non-negative, -Wsign-compare won't fire,
2806  // but we should still check for comparisons which are always true
2807  // or false.
2808  if (signedRange.NonNegative)
2809    return CheckTrivialUnsignedComparison(S, E);
2810
2811  // For (in)equality comparisons, if the unsigned operand is a
2812  // constant which cannot collide with a overflowed signed operand,
2813  // then reinterpreting the signed operand as unsigned will not
2814  // change the result of the comparison.
2815  if (E->isEqualityOp()) {
2816    unsigned comparisonWidth = S.Context.getIntWidth(T);
2817    IntRange unsignedRange = GetExprRange(S.Context, unsignedOperand);
2818
2819    // We should never be unable to prove that the unsigned operand is
2820    // non-negative.
2821    assert(unsignedRange.NonNegative && "unsigned range includes negative?");
2822
2823    if (unsignedRange.Width < comparisonWidth)
2824      return;
2825  }
2826
2827  S.Diag(E->getOperatorLoc(), diag::warn_mixed_sign_comparison)
2828    << lex->getType() << rex->getType()
2829    << lex->getSourceRange() << rex->getSourceRange();
2830}
2831
2832/// Analyzes an attempt to assign the given value to a bitfield.
2833///
2834/// Returns true if there was something fishy about the attempt.
2835bool AnalyzeBitFieldAssignment(Sema &S, FieldDecl *Bitfield, Expr *Init,
2836                               SourceLocation InitLoc) {
2837  assert(Bitfield->isBitField());
2838  if (Bitfield->isInvalidDecl())
2839    return false;
2840
2841  // White-list bool bitfields.
2842  if (Bitfield->getType()->isBooleanType())
2843    return false;
2844
2845  // Ignore value- or type-dependent expressions.
2846  if (Bitfield->getBitWidth()->isValueDependent() ||
2847      Bitfield->getBitWidth()->isTypeDependent() ||
2848      Init->isValueDependent() ||
2849      Init->isTypeDependent())
2850    return false;
2851
2852  Expr *OriginalInit = Init->IgnoreParenImpCasts();
2853
2854  llvm::APSInt Width(32);
2855  Expr::EvalResult InitValue;
2856  if (!Bitfield->getBitWidth()->isIntegerConstantExpr(Width, S.Context) ||
2857      !OriginalInit->Evaluate(InitValue, S.Context) ||
2858      !InitValue.Val.isInt())
2859    return false;
2860
2861  const llvm::APSInt &Value = InitValue.Val.getInt();
2862  unsigned OriginalWidth = Value.getBitWidth();
2863  unsigned FieldWidth = Width.getZExtValue();
2864
2865  if (OriginalWidth <= FieldWidth)
2866    return false;
2867
2868  llvm::APSInt TruncatedValue = Value.trunc(FieldWidth);
2869
2870  // It's fairly common to write values into signed bitfields
2871  // that, if sign-extended, would end up becoming a different
2872  // value.  We don't want to warn about that.
2873  if (Value.isSigned() && Value.isNegative())
2874    TruncatedValue = TruncatedValue.sext(OriginalWidth);
2875  else
2876    TruncatedValue = TruncatedValue.zext(OriginalWidth);
2877
2878  if (Value == TruncatedValue)
2879    return false;
2880
2881  std::string PrettyValue = Value.toString(10);
2882  std::string PrettyTrunc = TruncatedValue.toString(10);
2883
2884  S.Diag(InitLoc, diag::warn_impcast_bitfield_precision_constant)
2885    << PrettyValue << PrettyTrunc << OriginalInit->getType()
2886    << Init->getSourceRange();
2887
2888  return true;
2889}
2890
2891/// Analyze the given simple or compound assignment for warning-worthy
2892/// operations.
2893void AnalyzeAssignment(Sema &S, BinaryOperator *E) {
2894  // Just recurse on the LHS.
2895  AnalyzeImplicitConversions(S, E->getLHS(), E->getOperatorLoc());
2896
2897  // We want to recurse on the RHS as normal unless we're assigning to
2898  // a bitfield.
2899  if (FieldDecl *Bitfield = E->getLHS()->getBitField()) {
2900    if (AnalyzeBitFieldAssignment(S, Bitfield, E->getRHS(),
2901                                  E->getOperatorLoc())) {
2902      // Recurse, ignoring any implicit conversions on the RHS.
2903      return AnalyzeImplicitConversions(S, E->getRHS()->IgnoreParenImpCasts(),
2904                                        E->getOperatorLoc());
2905    }
2906  }
2907
2908  AnalyzeImplicitConversions(S, E->getRHS(), E->getOperatorLoc());
2909}
2910
2911/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
2912void DiagnoseImpCast(Sema &S, Expr *E, QualType SourceType, QualType T,
2913                     SourceLocation CContext, unsigned diag) {
2914  S.Diag(E->getExprLoc(), diag)
2915    << SourceType << T << E->getSourceRange() << SourceRange(CContext);
2916}
2917
2918/// Diagnose an implicit cast;  purely a helper for CheckImplicitConversion.
2919void DiagnoseImpCast(Sema &S, Expr *E, QualType T, SourceLocation CContext,
2920                     unsigned diag) {
2921  DiagnoseImpCast(S, E, E->getType(), T, CContext, diag);
2922}
2923
2924/// Diagnose an implicit cast from a literal expression. Also attemps to supply
2925/// fixit hints when the cast wouldn't lose information to simply write the
2926/// expression with the expected type.
2927void DiagnoseFloatingLiteralImpCast(Sema &S, FloatingLiteral *FL, QualType T,
2928                                    SourceLocation CContext) {
2929  // Emit the primary warning first, then try to emit a fixit hint note if
2930  // reasonable.
2931  S.Diag(FL->getExprLoc(), diag::warn_impcast_literal_float_to_integer)
2932    << FL->getType() << T << FL->getSourceRange() << SourceRange(CContext);
2933
2934  const llvm::APFloat &Value = FL->getValue();
2935
2936  // Don't attempt to fix PPC double double literals.
2937  if (&Value.getSemantics() == &llvm::APFloat::PPCDoubleDouble)
2938    return;
2939
2940  // Try to convert this exactly to an 64-bit integer. FIXME: It would be
2941  // nice to support arbitrarily large integers here.
2942  bool isExact = false;
2943  uint64_t IntegerPart;
2944  if (Value.convertToInteger(&IntegerPart, 64, /*isSigned=*/true,
2945                             llvm::APFloat::rmTowardZero, &isExact)
2946      != llvm::APFloat::opOK || !isExact)
2947    return;
2948
2949  llvm::APInt IntegerValue(64, IntegerPart, /*isSigned=*/true);
2950
2951  std::string LiteralValue = IntegerValue.toString(10, /*isSigned=*/true);
2952  S.Diag(FL->getExprLoc(), diag::note_fix_integral_float_as_integer)
2953    << FixItHint::CreateReplacement(FL->getSourceRange(), LiteralValue);
2954}
2955
2956std::string PrettyPrintInRange(const llvm::APSInt &Value, IntRange Range) {
2957  if (!Range.Width) return "0";
2958
2959  llvm::APSInt ValueInRange = Value;
2960  ValueInRange.setIsSigned(!Range.NonNegative);
2961  ValueInRange = ValueInRange.trunc(Range.Width);
2962  return ValueInRange.toString(10);
2963}
2964
2965static bool isFromSystemMacro(Sema &S, SourceLocation loc) {
2966  SourceManager &smgr = S.Context.getSourceManager();
2967  return loc.isMacroID() && smgr.isInSystemHeader(smgr.getSpellingLoc(loc));
2968}
2969
2970void CheckImplicitConversion(Sema &S, Expr *E, QualType T,
2971                             SourceLocation CC, bool *ICContext = 0) {
2972  if (E->isTypeDependent() || E->isValueDependent()) return;
2973
2974  const Type *Source = S.Context.getCanonicalType(E->getType()).getTypePtr();
2975  const Type *Target = S.Context.getCanonicalType(T).getTypePtr();
2976  if (Source == Target) return;
2977  if (Target->isDependentType()) return;
2978
2979  // If the conversion context location is invalid don't complain.
2980  // We also don't want to emit a warning if the issue occurs from the
2981  // instantiation of a system macro.  The problem is that 'getSpellingLoc()'
2982  // is slow, so we delay this check as long as possible.  Once we detect
2983  // we are in that scenario, we just return.
2984  if (CC.isInvalid())
2985    return;
2986
2987  // Never diagnose implicit casts to bool.
2988  if (Target->isSpecificBuiltinType(BuiltinType::Bool))
2989    return;
2990
2991  // Strip vector types.
2992  if (isa<VectorType>(Source)) {
2993    if (!isa<VectorType>(Target)) {
2994      if (isFromSystemMacro(S, CC))
2995        return;
2996      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_vector_scalar);
2997    }
2998
2999    // If the vector cast is cast between two vectors of the same size, it is
3000    // a bitcast, not a conversion.
3001    if (S.Context.getTypeSize(Source) == S.Context.getTypeSize(Target))
3002      return;
3003
3004    Source = cast<VectorType>(Source)->getElementType().getTypePtr();
3005    Target = cast<VectorType>(Target)->getElementType().getTypePtr();
3006  }
3007
3008  // Strip complex types.
3009  if (isa<ComplexType>(Source)) {
3010    if (!isa<ComplexType>(Target)) {
3011      if (isFromSystemMacro(S, CC))
3012        return;
3013
3014      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_complex_scalar);
3015    }
3016
3017    Source = cast<ComplexType>(Source)->getElementType().getTypePtr();
3018    Target = cast<ComplexType>(Target)->getElementType().getTypePtr();
3019  }
3020
3021  const BuiltinType *SourceBT = dyn_cast<BuiltinType>(Source);
3022  const BuiltinType *TargetBT = dyn_cast<BuiltinType>(Target);
3023
3024  // If the source is floating point...
3025  if (SourceBT && SourceBT->isFloatingPoint()) {
3026    // ...and the target is floating point...
3027    if (TargetBT && TargetBT->isFloatingPoint()) {
3028      // ...then warn if we're dropping FP rank.
3029
3030      // Builtin FP kinds are ordered by increasing FP rank.
3031      if (SourceBT->getKind() > TargetBT->getKind()) {
3032        // Don't warn about float constants that are precisely
3033        // representable in the target type.
3034        Expr::EvalResult result;
3035        if (E->Evaluate(result, S.Context)) {
3036          // Value might be a float, a float vector, or a float complex.
3037          if (IsSameFloatAfterCast(result.Val,
3038                   S.Context.getFloatTypeSemantics(QualType(TargetBT, 0)),
3039                   S.Context.getFloatTypeSemantics(QualType(SourceBT, 0))))
3040            return;
3041        }
3042
3043        if (isFromSystemMacro(S, CC))
3044          return;
3045
3046        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_precision);
3047      }
3048      return;
3049    }
3050
3051    // If the target is integral, always warn.
3052    if ((TargetBT && TargetBT->isInteger())) {
3053      if (isFromSystemMacro(S, CC))
3054        return;
3055
3056      Expr *InnerE = E->IgnoreParenImpCasts();
3057      if (FloatingLiteral *FL = dyn_cast<FloatingLiteral>(InnerE)) {
3058        DiagnoseFloatingLiteralImpCast(S, FL, T, CC);
3059      } else {
3060        DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_float_integer);
3061      }
3062    }
3063
3064    return;
3065  }
3066
3067  if (!Source->isIntegerType() || !Target->isIntegerType())
3068    return;
3069
3070  if ((E->isNullPointerConstant(S.Context, Expr::NPC_ValueDependentIsNotNull)
3071           == Expr::NPCK_GNUNull) && Target->isIntegerType()) {
3072    S.Diag(E->getExprLoc(), diag::warn_impcast_null_pointer_to_integer)
3073        << E->getSourceRange() << clang::SourceRange(CC);
3074    return;
3075  }
3076
3077  IntRange SourceRange = GetExprRange(S.Context, E);
3078  IntRange TargetRange = IntRange::forTargetOfCanonicalType(S.Context, Target);
3079
3080  if (SourceRange.Width > TargetRange.Width) {
3081    // If the source is a constant, use a default-on diagnostic.
3082    // TODO: this should happen for bitfield stores, too.
3083    llvm::APSInt Value(32);
3084    if (E->isIntegerConstantExpr(Value, S.Context)) {
3085      if (isFromSystemMacro(S, CC))
3086        return;
3087
3088      std::string PrettySourceValue = Value.toString(10);
3089      std::string PrettyTargetValue = PrettyPrintInRange(Value, TargetRange);
3090
3091      S.Diag(E->getExprLoc(), diag::warn_impcast_integer_precision_constant)
3092        << PrettySourceValue << PrettyTargetValue
3093        << E->getType() << T << E->getSourceRange() << clang::SourceRange(CC);
3094      return;
3095    }
3096
3097    // People want to build with -Wshorten-64-to-32 and not -Wconversion.
3098    if (isFromSystemMacro(S, CC))
3099      return;
3100
3101    if (SourceRange.Width == 64 && TargetRange.Width == 32)
3102      return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_64_32);
3103    return DiagnoseImpCast(S, E, T, CC, diag::warn_impcast_integer_precision);
3104  }
3105
3106  if ((TargetRange.NonNegative && !SourceRange.NonNegative) ||
3107      (!TargetRange.NonNegative && SourceRange.NonNegative &&
3108       SourceRange.Width == TargetRange.Width)) {
3109
3110    if (isFromSystemMacro(S, CC))
3111      return;
3112
3113    unsigned DiagID = diag::warn_impcast_integer_sign;
3114
3115    // Traditionally, gcc has warned about this under -Wsign-compare.
3116    // We also want to warn about it in -Wconversion.
3117    // So if -Wconversion is off, use a completely identical diagnostic
3118    // in the sign-compare group.
3119    // The conditional-checking code will
3120    if (ICContext) {
3121      DiagID = diag::warn_impcast_integer_sign_conditional;
3122      *ICContext = true;
3123    }
3124
3125    return DiagnoseImpCast(S, E, T, CC, DiagID);
3126  }
3127
3128  // Diagnose conversions between different enumeration types.
3129  // In C, we pretend that the type of an EnumConstantDecl is its enumeration
3130  // type, to give us better diagnostics.
3131  QualType SourceType = E->getType();
3132  if (!S.getLangOptions().CPlusPlus) {
3133    if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
3134      if (EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(DRE->getDecl())) {
3135        EnumDecl *Enum = cast<EnumDecl>(ECD->getDeclContext());
3136        SourceType = S.Context.getTypeDeclType(Enum);
3137        Source = S.Context.getCanonicalType(SourceType).getTypePtr();
3138      }
3139  }
3140
3141  if (const EnumType *SourceEnum = Source->getAs<EnumType>())
3142    if (const EnumType *TargetEnum = Target->getAs<EnumType>())
3143      if ((SourceEnum->getDecl()->getIdentifier() ||
3144           SourceEnum->getDecl()->getTypedefNameForAnonDecl()) &&
3145          (TargetEnum->getDecl()->getIdentifier() ||
3146           TargetEnum->getDecl()->getTypedefNameForAnonDecl()) &&
3147          SourceEnum != TargetEnum) {
3148        if (isFromSystemMacro(S, CC))
3149          return;
3150
3151        return DiagnoseImpCast(S, E, SourceType, T, CC,
3152                               diag::warn_impcast_different_enum_types);
3153      }
3154
3155  return;
3156}
3157
3158void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T);
3159
3160void CheckConditionalOperand(Sema &S, Expr *E, QualType T,
3161                             SourceLocation CC, bool &ICContext) {
3162  E = E->IgnoreParenImpCasts();
3163
3164  if (isa<ConditionalOperator>(E))
3165    return CheckConditionalOperator(S, cast<ConditionalOperator>(E), T);
3166
3167  AnalyzeImplicitConversions(S, E, CC);
3168  if (E->getType() != T)
3169    return CheckImplicitConversion(S, E, T, CC, &ICContext);
3170  return;
3171}
3172
3173void CheckConditionalOperator(Sema &S, ConditionalOperator *E, QualType T) {
3174  SourceLocation CC = E->getQuestionLoc();
3175
3176  AnalyzeImplicitConversions(S, E->getCond(), CC);
3177
3178  bool Suspicious = false;
3179  CheckConditionalOperand(S, E->getTrueExpr(), T, CC, Suspicious);
3180  CheckConditionalOperand(S, E->getFalseExpr(), T, CC, Suspicious);
3181
3182  // If -Wconversion would have warned about either of the candidates
3183  // for a signedness conversion to the context type...
3184  if (!Suspicious) return;
3185
3186  // ...but it's currently ignored...
3187  if (S.Diags.getDiagnosticLevel(diag::warn_impcast_integer_sign_conditional,
3188                                 CC))
3189    return;
3190
3191  // ...and -Wsign-compare isn't...
3192  if (!S.Diags.getDiagnosticLevel(diag::warn_mixed_sign_conditional, CC))
3193    return;
3194
3195  // ...then check whether it would have warned about either of the
3196  // candidates for a signedness conversion to the condition type.
3197  if (E->getType() != T) {
3198    Suspicious = false;
3199    CheckImplicitConversion(S, E->getTrueExpr()->IgnoreParenImpCasts(),
3200                            E->getType(), CC, &Suspicious);
3201    if (!Suspicious)
3202      CheckImplicitConversion(S, E->getFalseExpr()->IgnoreParenImpCasts(),
3203                              E->getType(), CC, &Suspicious);
3204    if (!Suspicious)
3205      return;
3206  }
3207
3208  // If so, emit a diagnostic under -Wsign-compare.
3209  Expr *lex = E->getTrueExpr()->IgnoreParenImpCasts();
3210  Expr *rex = E->getFalseExpr()->IgnoreParenImpCasts();
3211  S.Diag(E->getQuestionLoc(), diag::warn_mixed_sign_conditional)
3212    << lex->getType() << rex->getType()
3213    << lex->getSourceRange() << rex->getSourceRange();
3214}
3215
3216/// AnalyzeImplicitConversions - Find and report any interesting
3217/// implicit conversions in the given expression.  There are a couple
3218/// of competing diagnostics here, -Wconversion and -Wsign-compare.
3219void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) {
3220  QualType T = OrigE->getType();
3221  Expr *E = OrigE->IgnoreParenImpCasts();
3222
3223  // For conditional operators, we analyze the arguments as if they
3224  // were being fed directly into the output.
3225  if (isa<ConditionalOperator>(E)) {
3226    ConditionalOperator *CO = cast<ConditionalOperator>(E);
3227    CheckConditionalOperator(S, CO, T);
3228    return;
3229  }
3230
3231  // Go ahead and check any implicit conversions we might have skipped.
3232  // The non-canonical typecheck is just an optimization;
3233  // CheckImplicitConversion will filter out dead implicit conversions.
3234  if (E->getType() != T)
3235    CheckImplicitConversion(S, E, T, CC);
3236
3237  // Now continue drilling into this expression.
3238
3239  // Skip past explicit casts.
3240  if (isa<ExplicitCastExpr>(E)) {
3241    E = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreParenImpCasts();
3242    return AnalyzeImplicitConversions(S, E, CC);
3243  }
3244
3245  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
3246    // Do a somewhat different check with comparison operators.
3247    if (BO->isComparisonOp())
3248      return AnalyzeComparison(S, BO);
3249
3250    // And with assignments and compound assignments.
3251    if (BO->isAssignmentOp())
3252      return AnalyzeAssignment(S, BO);
3253  }
3254
3255  // These break the otherwise-useful invariant below.  Fortunately,
3256  // we don't really need to recurse into them, because any internal
3257  // expressions should have been analyzed already when they were
3258  // built into statements.
3259  if (isa<StmtExpr>(E)) return;
3260
3261  // Don't descend into unevaluated contexts.
3262  if (isa<UnaryExprOrTypeTraitExpr>(E)) return;
3263
3264  // Now just recurse over the expression's children.
3265  CC = E->getExprLoc();
3266  for (Stmt::child_range I = E->children(); I; ++I)
3267    AnalyzeImplicitConversions(S, cast<Expr>(*I), CC);
3268}
3269
3270} // end anonymous namespace
3271
3272/// Diagnoses "dangerous" implicit conversions within the given
3273/// expression (which is a full expression).  Implements -Wconversion
3274/// and -Wsign-compare.
3275///
3276/// \param CC the "context" location of the implicit conversion, i.e.
3277///   the most location of the syntactic entity requiring the implicit
3278///   conversion
3279void Sema::CheckImplicitConversions(Expr *E, SourceLocation CC) {
3280  // Don't diagnose in unevaluated contexts.
3281  if (ExprEvalContexts.back().Context == Sema::Unevaluated)
3282    return;
3283
3284  // Don't diagnose for value- or type-dependent expressions.
3285  if (E->isTypeDependent() || E->isValueDependent())
3286    return;
3287
3288  // This is not the right CC for (e.g.) a variable initialization.
3289  AnalyzeImplicitConversions(*this, E, CC);
3290}
3291
3292void Sema::CheckBitFieldInitialization(SourceLocation InitLoc,
3293                                       FieldDecl *BitField,
3294                                       Expr *Init) {
3295  (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc);
3296}
3297
3298/// CheckParmsForFunctionDef - Check that the parameters of the given
3299/// function are appropriate for the definition of a function. This
3300/// takes care of any checks that cannot be performed on the
3301/// declaration itself, e.g., that the types of each of the function
3302/// parameters are complete.
3303bool Sema::CheckParmsForFunctionDef(ParmVarDecl **P, ParmVarDecl **PEnd,
3304                                    bool CheckParameterNames) {
3305  bool HasInvalidParm = false;
3306  for (; P != PEnd; ++P) {
3307    ParmVarDecl *Param = *P;
3308
3309    // C99 6.7.5.3p4: the parameters in a parameter type list in a
3310    // function declarator that is part of a function definition of
3311    // that function shall not have incomplete type.
3312    //
3313    // This is also C++ [dcl.fct]p6.
3314    if (!Param->isInvalidDecl() &&
3315        RequireCompleteType(Param->getLocation(), Param->getType(),
3316                               diag::err_typecheck_decl_incomplete_type)) {
3317      Param->setInvalidDecl();
3318      HasInvalidParm = true;
3319    }
3320
3321    // C99 6.9.1p5: If the declarator includes a parameter type list, the
3322    // declaration of each parameter shall include an identifier.
3323    if (CheckParameterNames &&
3324        Param->getIdentifier() == 0 &&
3325        !Param->isImplicit() &&
3326        !getLangOptions().CPlusPlus)
3327      Diag(Param->getLocation(), diag::err_parameter_name_omitted);
3328
3329    // C99 6.7.5.3p12:
3330    //   If the function declarator is not part of a definition of that
3331    //   function, parameters may have incomplete type and may use the [*]
3332    //   notation in their sequences of declarator specifiers to specify
3333    //   variable length array types.
3334    QualType PType = Param->getOriginalType();
3335    if (const ArrayType *AT = Context.getAsArrayType(PType)) {
3336      if (AT->getSizeModifier() == ArrayType::Star) {
3337        // FIXME: This diagnosic should point the the '[*]' if source-location
3338        // information is added for it.
3339        Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
3340      }
3341    }
3342  }
3343
3344  return HasInvalidParm;
3345}
3346
3347/// CheckCastAlign - Implements -Wcast-align, which warns when a
3348/// pointer cast increases the alignment requirements.
3349void Sema::CheckCastAlign(Expr *Op, QualType T, SourceRange TRange) {
3350  // This is actually a lot of work to potentially be doing on every
3351  // cast; don't do it if we're ignoring -Wcast_align (as is the default).
3352  if (getDiagnostics().getDiagnosticLevel(diag::warn_cast_align,
3353                                          TRange.getBegin())
3354        == Diagnostic::Ignored)
3355    return;
3356
3357  // Ignore dependent types.
3358  if (T->isDependentType() || Op->getType()->isDependentType())
3359    return;
3360
3361  // Require that the destination be a pointer type.
3362  const PointerType *DestPtr = T->getAs<PointerType>();
3363  if (!DestPtr) return;
3364
3365  // If the destination has alignment 1, we're done.
3366  QualType DestPointee = DestPtr->getPointeeType();
3367  if (DestPointee->isIncompleteType()) return;
3368  CharUnits DestAlign = Context.getTypeAlignInChars(DestPointee);
3369  if (DestAlign.isOne()) return;
3370
3371  // Require that the source be a pointer type.
3372  const PointerType *SrcPtr = Op->getType()->getAs<PointerType>();
3373  if (!SrcPtr) return;
3374  QualType SrcPointee = SrcPtr->getPointeeType();
3375
3376  // Whitelist casts from cv void*.  We already implicitly
3377  // whitelisted casts to cv void*, since they have alignment 1.
3378  // Also whitelist casts involving incomplete types, which implicitly
3379  // includes 'void'.
3380  if (SrcPointee->isIncompleteType()) return;
3381
3382  CharUnits SrcAlign = Context.getTypeAlignInChars(SrcPointee);
3383  if (SrcAlign >= DestAlign) return;
3384
3385  Diag(TRange.getBegin(), diag::warn_cast_align)
3386    << Op->getType() << T
3387    << static_cast<unsigned>(SrcAlign.getQuantity())
3388    << static_cast<unsigned>(DestAlign.getQuantity())
3389    << TRange << Op->getSourceRange();
3390}
3391
3392static void CheckArrayAccess_Check(Sema &S,
3393                                   const clang::ArraySubscriptExpr *E) {
3394  const Expr *BaseExpr = E->getBase()->IgnoreParenImpCasts();
3395  const ConstantArrayType *ArrayTy =
3396    S.Context.getAsConstantArrayType(BaseExpr->getType());
3397  if (!ArrayTy)
3398    return;
3399
3400  const Expr *IndexExpr = E->getIdx();
3401  if (IndexExpr->isValueDependent())
3402    return;
3403  llvm::APSInt index;
3404  if (!IndexExpr->isIntegerConstantExpr(index, S.Context))
3405    return;
3406
3407  if (index.isUnsigned() || !index.isNegative()) {
3408    llvm::APInt size = ArrayTy->getSize();
3409    if (!size.isStrictlyPositive())
3410      return;
3411    if (size.getBitWidth() > index.getBitWidth())
3412      index = index.sext(size.getBitWidth());
3413    else if (size.getBitWidth() < index.getBitWidth())
3414      size = size.sext(index.getBitWidth());
3415
3416    if (index.slt(size))
3417      return;
3418
3419    S.DiagRuntimeBehavior(E->getBase()->getLocStart(), BaseExpr,
3420                          S.PDiag(diag::warn_array_index_exceeds_bounds)
3421                            << index.toString(10, true)
3422                            << size.toString(10, true)
3423                            << IndexExpr->getSourceRange());
3424  } else {
3425    S.DiagRuntimeBehavior(E->getBase()->getLocStart(), BaseExpr,
3426                          S.PDiag(diag::warn_array_index_precedes_bounds)
3427                            << index.toString(10, true)
3428                            << IndexExpr->getSourceRange());
3429  }
3430
3431  const NamedDecl *ND = NULL;
3432  if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(BaseExpr))
3433    ND = dyn_cast<NamedDecl>(DRE->getDecl());
3434  if (const MemberExpr *ME = dyn_cast<MemberExpr>(BaseExpr))
3435    ND = dyn_cast<NamedDecl>(ME->getMemberDecl());
3436  if (ND)
3437    S.DiagRuntimeBehavior(ND->getLocStart(), BaseExpr,
3438                          S.PDiag(diag::note_array_index_out_of_bounds)
3439                            << ND->getDeclName());
3440}
3441
3442void Sema::CheckArrayAccess(const Expr *expr) {
3443  while (true) {
3444    expr = expr->IgnoreParens();
3445    switch (expr->getStmtClass()) {
3446      case Stmt::ArraySubscriptExprClass:
3447        CheckArrayAccess_Check(*this, cast<ArraySubscriptExpr>(expr));
3448        return;
3449      case Stmt::ConditionalOperatorClass: {
3450        const ConditionalOperator *cond = cast<ConditionalOperator>(expr);
3451        if (const Expr *lhs = cond->getLHS())
3452          CheckArrayAccess(lhs);
3453        if (const Expr *rhs = cond->getRHS())
3454          CheckArrayAccess(rhs);
3455        return;
3456      }
3457      default:
3458        return;
3459    }
3460  }
3461}
3462
3463//===--- CHECK: Objective-C retain cycles ----------------------------------//
3464
3465namespace {
3466  struct RetainCycleOwner {
3467    RetainCycleOwner() : Variable(0), Indirect(false) {}
3468    VarDecl *Variable;
3469    SourceRange Range;
3470    SourceLocation Loc;
3471    bool Indirect;
3472
3473    void setLocsFrom(Expr *e) {
3474      Loc = e->getExprLoc();
3475      Range = e->getSourceRange();
3476    }
3477  };
3478}
3479
3480/// Consider whether capturing the given variable can possibly lead to
3481/// a retain cycle.
3482static bool considerVariable(VarDecl *var, Expr *ref, RetainCycleOwner &owner) {
3483  // In ARC, it's captured strongly iff the variable has __strong
3484  // lifetime.  In MRR, it's captured strongly if the variable is
3485  // __block and has an appropriate type.
3486  if (var->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
3487    return false;
3488
3489  owner.Variable = var;
3490  owner.setLocsFrom(ref);
3491  return true;
3492}
3493
3494static bool findRetainCycleOwner(Expr *e, RetainCycleOwner &owner) {
3495  while (true) {
3496    e = e->IgnoreParens();
3497    if (CastExpr *cast = dyn_cast<CastExpr>(e)) {
3498      switch (cast->getCastKind()) {
3499      case CK_BitCast:
3500      case CK_LValueBitCast:
3501      case CK_LValueToRValue:
3502        e = cast->getSubExpr();
3503        continue;
3504
3505      case CK_GetObjCProperty: {
3506        // Bail out if this isn't a strong explicit property.
3507        const ObjCPropertyRefExpr *pre = cast->getSubExpr()->getObjCProperty();
3508        if (pre->isImplicitProperty()) return false;
3509        ObjCPropertyDecl *property = pre->getExplicitProperty();
3510        if (!(property->getPropertyAttributes() &
3511              (ObjCPropertyDecl::OBJC_PR_retain |
3512               ObjCPropertyDecl::OBJC_PR_copy |
3513               ObjCPropertyDecl::OBJC_PR_strong)) &&
3514            !(property->getPropertyIvarDecl() &&
3515              property->getPropertyIvarDecl()->getType()
3516                .getObjCLifetime() == Qualifiers::OCL_Strong))
3517          return false;
3518
3519        owner.Indirect = true;
3520        e = const_cast<Expr*>(pre->getBase());
3521        continue;
3522      }
3523
3524      default:
3525        return false;
3526      }
3527    }
3528
3529    if (ObjCIvarRefExpr *ref = dyn_cast<ObjCIvarRefExpr>(e)) {
3530      ObjCIvarDecl *ivar = ref->getDecl();
3531      if (ivar->getType().getObjCLifetime() != Qualifiers::OCL_Strong)
3532        return false;
3533
3534      // Try to find a retain cycle in the base.
3535      if (!findRetainCycleOwner(ref->getBase(), owner))
3536        return false;
3537
3538      if (ref->isFreeIvar()) owner.setLocsFrom(ref);
3539      owner.Indirect = true;
3540      return true;
3541    }
3542
3543    if (DeclRefExpr *ref = dyn_cast<DeclRefExpr>(e)) {
3544      VarDecl *var = dyn_cast<VarDecl>(ref->getDecl());
3545      if (!var) return false;
3546      return considerVariable(var, ref, owner);
3547    }
3548
3549    if (BlockDeclRefExpr *ref = dyn_cast<BlockDeclRefExpr>(e)) {
3550      owner.Variable = ref->getDecl();
3551      owner.setLocsFrom(ref);
3552      return true;
3553    }
3554
3555    if (MemberExpr *member = dyn_cast<MemberExpr>(e)) {
3556      if (member->isArrow()) return false;
3557
3558      // Don't count this as an indirect ownership.
3559      e = member->getBase();
3560      continue;
3561    }
3562
3563    // Array ivars?
3564
3565    return false;
3566  }
3567}
3568
3569namespace {
3570  struct FindCaptureVisitor : EvaluatedExprVisitor<FindCaptureVisitor> {
3571    FindCaptureVisitor(ASTContext &Context, VarDecl *variable)
3572      : EvaluatedExprVisitor<FindCaptureVisitor>(Context),
3573        Variable(variable), Capturer(0) {}
3574
3575    VarDecl *Variable;
3576    Expr *Capturer;
3577
3578    void VisitDeclRefExpr(DeclRefExpr *ref) {
3579      if (ref->getDecl() == Variable && !Capturer)
3580        Capturer = ref;
3581    }
3582
3583    void VisitBlockDeclRefExpr(BlockDeclRefExpr *ref) {
3584      if (ref->getDecl() == Variable && !Capturer)
3585        Capturer = ref;
3586    }
3587
3588    void VisitObjCIvarRefExpr(ObjCIvarRefExpr *ref) {
3589      if (Capturer) return;
3590      Visit(ref->getBase());
3591      if (Capturer && ref->isFreeIvar())
3592        Capturer = ref;
3593    }
3594
3595    void VisitBlockExpr(BlockExpr *block) {
3596      // Look inside nested blocks
3597      if (block->getBlockDecl()->capturesVariable(Variable))
3598        Visit(block->getBlockDecl()->getBody());
3599    }
3600  };
3601}
3602
3603/// Check whether the given argument is a block which captures a
3604/// variable.
3605static Expr *findCapturingExpr(Sema &S, Expr *e, RetainCycleOwner &owner) {
3606  assert(owner.Variable && owner.Loc.isValid());
3607
3608  e = e->IgnoreParenCasts();
3609  BlockExpr *block = dyn_cast<BlockExpr>(e);
3610  if (!block || !block->getBlockDecl()->capturesVariable(owner.Variable))
3611    return 0;
3612
3613  FindCaptureVisitor visitor(S.Context, owner.Variable);
3614  visitor.Visit(block->getBlockDecl()->getBody());
3615  return visitor.Capturer;
3616}
3617
3618static void diagnoseRetainCycle(Sema &S, Expr *capturer,
3619                                RetainCycleOwner &owner) {
3620  assert(capturer);
3621  assert(owner.Variable && owner.Loc.isValid());
3622
3623  S.Diag(capturer->getExprLoc(), diag::warn_arc_retain_cycle)
3624    << owner.Variable << capturer->getSourceRange();
3625  S.Diag(owner.Loc, diag::note_arc_retain_cycle_owner)
3626    << owner.Indirect << owner.Range;
3627}
3628
3629/// Check for a keyword selector that starts with the word 'add' or
3630/// 'set'.
3631static bool isSetterLikeSelector(Selector sel) {
3632  if (sel.isUnarySelector()) return false;
3633
3634  llvm::StringRef str = sel.getNameForSlot(0);
3635  while (!str.empty() && str.front() == '_') str = str.substr(1);
3636  if (str.startswith("set") || str.startswith("add"))
3637    str = str.substr(3);
3638  else
3639    return false;
3640
3641  if (str.empty()) return true;
3642  return !islower(str.front());
3643}
3644
3645/// Check a message send to see if it's likely to cause a retain cycle.
3646void Sema::checkRetainCycles(ObjCMessageExpr *msg) {
3647  // Only check instance methods whose selector looks like a setter.
3648  if (!msg->isInstanceMessage() || !isSetterLikeSelector(msg->getSelector()))
3649    return;
3650
3651  // Try to find a variable that the receiver is strongly owned by.
3652  RetainCycleOwner owner;
3653  if (msg->getReceiverKind() == ObjCMessageExpr::Instance) {
3654    if (!findRetainCycleOwner(msg->getInstanceReceiver(), owner))
3655      return;
3656  } else {
3657    assert(msg->getReceiverKind() == ObjCMessageExpr::SuperInstance);
3658    owner.Variable = getCurMethodDecl()->getSelfDecl();
3659    owner.Loc = msg->getSuperLoc();
3660    owner.Range = msg->getSuperLoc();
3661  }
3662
3663  // Check whether the receiver is captured by any of the arguments.
3664  for (unsigned i = 0, e = msg->getNumArgs(); i != e; ++i)
3665    if (Expr *capturer = findCapturingExpr(*this, msg->getArg(i), owner))
3666      return diagnoseRetainCycle(*this, capturer, owner);
3667}
3668
3669/// Check a property assign to see if it's likely to cause a retain cycle.
3670void Sema::checkRetainCycles(Expr *receiver, Expr *argument) {
3671  RetainCycleOwner owner;
3672  if (!findRetainCycleOwner(receiver, owner))
3673    return;
3674
3675  if (Expr *capturer = findCapturingExpr(*this, argument, owner))
3676    diagnoseRetainCycle(*this, capturer, owner);
3677}
3678
3679void Sema::checkUnsafeAssigns(SourceLocation Loc,
3680                              QualType LHS, Expr *RHS) {
3681  Qualifiers::ObjCLifetime LT = LHS.getObjCLifetime();
3682  if (LT != Qualifiers::OCL_Weak && LT != Qualifiers::OCL_ExplicitNone)
3683    return;
3684  if (ImplicitCastExpr *cast = dyn_cast<ImplicitCastExpr>(RHS))
3685    if (cast->getCastKind() == CK_ObjCConsumeObject)
3686      Diag(Loc, diag::warn_arc_retained_assign)
3687        << (LT == Qualifiers::OCL_ExplicitNone)
3688        << RHS->getSourceRange();
3689}
3690
3691