[clang] [llvm] Enable fexec-charset option (PR #138895)

[Sergei Barannikov via cfe-commits]

================
@@ -2332,12 +2422,29 @@ static const char *resyncUTF8(const char *Err, const char *End) {
 /// This function copies from Fragment, which is a sequence of bytes
 /// within Tok's contents (which begin at TokBegin) into ResultPtr.
 /// Performs widening for multi-byte characters.
-bool StringLiteralParser::CopyStringFragment(const Token &Tok,
-                                             const char *TokBegin,
-                                             StringRef Fragment) {
+bool StringLiteralParser::CopyStringFragment(
+    const Token &Tok, const char *TokBegin, StringRef Fragment,
+    llvm::TextEncodingConverter *Converter) {
+
   const llvm::UTF8 *ErrorPtrTmp;
-  if (ConvertUTF8toWide(CharByteWidth, Fragment, ResultPtr, ErrorPtrTmp))
+  if (ConvertUTF8toWide(CharByteWidth, Fragment, ResultPtr, ErrorPtrTmp)) {
+    if (Converter) {
+      assert(isOrdinary() && "Only ordinary literals are supported");
+      SmallString<64> CpConv;
+      char *Cp = ResultPtr - Fragment.size();
+      auto EC = Converter->convert(Fragment, CpConv);
+      if (!EC) {
+        memcpy(Cp, CpConv.data(), CpConv.size());
+        ResultPtr = Cp + CpConv.size();
+      } else { // there was a conversion error
+        if (Diags)
+          Diags->Report(Tok.getLocation(),
+                        diag::err_exec_charset_conversion_failed)
+              << EC.message();
+      }
+    }
----------------
s-barannikov wrote:

Although ConvertUTF8toWide can handle CharByteWidth==1 (by copying the input to the ouptut), the name suggests it should actually *convert* a string. We should either do this conversion inside that function (and hopefully give it a more precise name), or move this logic above the call (duplicating isLegalUTF8String check).


https://github.com/llvm/llvm-project/pull/138895