[clang] [clang] implement current direction of CWG2765 for string literal comparisons in constant evaluation (PR #109208)

[Timm Baeder via cfe-commits]

================
@@ -2142,11 +2150,91 @@ static const ValueDecl *GetLValueBaseDecl(const LValue &LVal) {
   return LVal.Base.dyn_cast<const ValueDecl*>();
 }
 
-static bool IsLiteralLValue(const LValue &Value) {
-  if (Value.getLValueCallIndex())
+// Information about an LValueBase that is some kind of string.
+struct LValueBaseString {
+  std::string ObjCEncodeStorage;
+  StringRef Bytes;
+  int CharWidth;
+};
+
+// Gets the lvalue base of LVal as a string.
+static bool GetLValueBaseAsString(const EvalInfo &Info, const LValue &LVal,
+                                  LValueBaseString &AsString) {
+  const auto *BaseExpr = LVal.Base.dyn_cast<const Expr *>();
+  if (!BaseExpr)
+    return false;
+
+  // For ObjCEncodeExpr, we need to compute and store the string.
+  if (const auto *EE = dyn_cast<ObjCEncodeExpr>(BaseExpr)) {
+    Info.Ctx.getObjCEncodingForType(EE->getEncodedType(),
+                                    AsString.ObjCEncodeStorage);
+    AsString.Bytes = AsString.ObjCEncodeStorage;
+    AsString.CharWidth = 1;
+    return true;
+  }
+
+  // Otherwise, we have a StringLiteral.
+  const auto *Lit = dyn_cast<StringLiteral>(BaseExpr);
+  if (const auto *PE = dyn_cast<PredefinedExpr>(BaseExpr))
+    Lit = PE->getFunctionName();
+
+  if (!Lit)
     return false;
-  const Expr *E = Value.Base.dyn_cast<const Expr*>();
-  return E && !isa<MaterializeTemporaryExpr>(E);
+
+  AsString.Bytes = Lit->getBytes();
+  AsString.CharWidth = Lit->getCharByteWidth();
+  return true;
+}
+
+// Determine whether two string literals potentially overlap. This will be the
+// case if they agree on the values of all the bytes on the overlapping region
+// between them.
+//
+// The overlapping region is the portion of the two string literals that must
+// overlap in memory if the pointers actually point to the same address at
+// runtime. For example, if LHS is "abcdef" + 3 and RHS is "cdef\0gh" + 1 then
+// the overlapping region is "cdef\0", which in this case does agree, so the
+// strings are potentially overlapping. Conversely, for "foobar" + 3 versus
+// "bazbar" + 3, the overlapping region contains all of both strings, so they
+// are not potentially overlapping, even though they agree from the given
+// addresses onwards.
+//
+// See open core issue CWG2765 which is discussing the desired rule here.
+static bool ArePotentiallyOverlappingStringLiterals(const EvalInfo &Info,
+                                                    const LValue &LHS,
+                                                    const LValue &RHS) {
+  LValueBaseString LHSString, RHSString;
+  if (!GetLValueBaseAsString(Info, LHS, LHSString) ||
+      !GetLValueBaseAsString(Info, RHS, RHSString))
+    return false;
+
+  // This is the byte offset to the location of the first character of LHS
+  // within RHS. We don't need to look at the characters of one string that
+  // would appear before the start of the other string if they were merged.
+  CharUnits Offset = RHS.Offset - LHS.Offset;
+  if (Offset.isNegative())
+    LHSString.Bytes = LHSString.Bytes.drop_front(-Offset.getQuantity());
+  else
+    RHSString.Bytes = RHSString.Bytes.drop_front(Offset.getQuantity());
+
+  bool LHSIsLonger = LHSString.Bytes.size() > RHSString.Bytes.size();
+  StringRef Longer = LHSIsLonger ? LHSString.Bytes : RHSString.Bytes;
+  StringRef Shorter = LHSIsLonger ? RHSString.Bytes : LHSString.Bytes;
+  int ShorterCharWidth = (LHSIsLonger ? RHSString : LHSString).CharWidth;
+
+  // The null terminator isn't included in the string data, so check for it
+  // manually. If the longer string doesn't have a null terminator where the
+  // shorter string ends, they aren't potentially overlapping.
+  for (int nullByte : llvm::seq(ShorterCharWidth)) {
----------------
tbaederr wrote:

```suggestion
  for (int NullByte : llvm::seq(ShorterCharWidth)) {
```

https://github.com/llvm/llvm-project/pull/109208