This change refactors some of the low-level lexer interfaces a bit.

Token now has a class of kinds for "literals", which include 
numeric constants, strings, etc.  These tokens can optionally have
a pointer to the start of the token in the lexer buffer.  This 
makes it faster to get spelling and do other gymnastics, because we
don't have to go through source locations.

This change is performance neutral, but will make other changes
more feasible down the road.



git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@63028 91177308-0d34-0410-b5e6-96231b3b80d8

Driver/CacheTokens.cpp

@@ -140,32 +140,19 @@ void PTHWriter::EmitToken(const Token& T) {
   Emit32(ResolveID(T.getIdentifierInfo()));
   Emit32(fpos);
   
-  // For specific tokens we cache their spelling.
-  if (T.getIdentifierInfo())
-    return;
+  // Literals (strings, numbers, characters) get cached spellings.
+  if (T.isLiteral()) {
+    // FIXME: This uses the slow getSpelling().  Perhaps we do better
+    // in the future?  This only slows down PTH generation.
+    const std::string &spelling = PP.getSpelling(T);
+    const char* s = spelling.c_str();
+    
+    // Get the string entry.
+    llvm::StringMapEntry<Offset> *E =
+      &CachedStrs.GetOrCreateValue(s, s+spelling.size());
 
-  switch (T.getKind()) {
-    default:
-      break;
-    case tok::string_literal:     
-    case tok::wide_string_literal:
-    case tok::angle_string_literal:
-    case tok::numeric_constant:
-    case tok::char_constant: {
-      // FIXME: This uses the slow getSpelling().  Perhaps we do better
-      // in the future?  This only slows down PTH generation.
-      const std::string& spelling = PP.getSpelling(T);
-      const char* s = spelling.c_str();
-      
-      // Get the string entry.
-      llvm::StringMapEntry<Offset> *E =
-        &CachedStrs.GetOrCreateValue(s, s+spelling.size());
-
-      // Store the address of the string entry in our spelling map.
-      (*CurSpellMap).push_back(std::make_pair(fpos, E));
-
-      break;
-    }
+    // Store the address of the string entry in our spelling map.
+    CurSpellMap->push_back(std::make_pair(fpos, E));
   }
 }
 

Driver/PrintPreprocessedOutput.cpp

@@ -430,6 +430,7 @@ bool PrintPPOutputPPCallbacks::AvoidConcat(const Token &PrevTok,
     // Avoid spelling identifiers, the most common form of token.
     FirstChar = II->getName()[0];
   } else if (!Tok.needsCleaning()) {
+    // FIXME: SPEED UP LITERALS!
     SourceManager &SrcMgr = PP.getSourceManager();
     FirstChar =
       *SrcMgr.getCharacterData(SrcMgr.getSpellingLoc(Tok.getLocation()));
@@ -556,6 +557,7 @@ void clang::DoPrintPreprocessedInput(Preprocessor &PP,
       const char *Str = II->getName();
       unsigned Len = Tok.needsCleaning() ? strlen(Str) : Tok.getLength();
       OS.write(Str, Len);
+      // FIXME: ACCELERATE LITERALS
     } else if (Tok.getLength() < 256) {
       const char *TokPtr = Buffer;
       unsigned Len = PP.getSpelling(Tok, TokPtr);

include/clang/Lex/Preprocessor.h

@@ -415,7 +415,7 @@ public:
   /// cached tokens doesn't get re-parsed and re-resolved after a backtrack is
   /// invoked.
   void AnnotateCachedTokens(const Token &Tok) {
-    assert(Tok.isAnnotationToken() && "Expected annotation token");
+    assert(Tok.isAnnotation() && "Expected annotation token");
     if (CachedLexPos != 0 && isBacktrackEnabled())
       AnnotatePreviousCachedTokens(Tok);
   }
@@ -463,11 +463,11 @@ public:
     return *SourceMgr.getCharacterData(SL);
   }
   
-  /// CreateString - Plop the specified string into a scratch buffer and return
-  /// a location for it.  If specified, the source location provides a source
-  /// location for the token.
-  SourceLocation CreateString(const char *Buf, unsigned Len,
-                              SourceLocation SourceLoc = SourceLocation());
+  /// CreateString - Plop the specified string into a scratch buffer and set the
+  /// specified token's location and length to it.  If specified, the source
+  /// location provides a location of the instantiation point of the token.
+  void CreateString(const char *Buf, unsigned Len,
+                    Token &Tok, SourceLocation SourceLoc = SourceLocation());
   
   /// DumpToken - Print the token to stderr, used for debugging.
   ///

include/clang/Lex/ScratchBuffer.h

@@ -30,17 +30,11 @@ class ScratchBuffer {
 public:
   ScratchBuffer(SourceManager &SM);
   
-  /// getToken - Splat the specified text into a temporary MemoryBuffer and
-  /// return a SourceLocation that refers to the token.  The SourceLoc value
-  /// gives a virtual location that the token will appear to be from.
-  SourceLocation getToken(const char *Buf, unsigned Len,
-                          SourceLocation SourceLoc);
-  
   /// getToken - Splat the specified text into a temporary MemoryBuffer and
   /// return a SourceLocation that refers to the token.  This is just like the
   /// previous method, but returns a location that indicates the physloc of the
   /// token.
-  SourceLocation getToken(const char *Buf, unsigned Len);
+  SourceLocation getToken(const char *Buf, unsigned Len, const char *&DestPtr);
                           
 private:
   void AllocScratchBuffer(unsigned RequestLen);

include/clang/Lex/Token.h

@@ -34,21 +34,28 @@ class Token {
   /// The location of the token.
   SourceLocation Loc;
 
-  // Conceptually these next two fields could be in a union with
-  // access depending on isAnnotationToken(). However, this causes gcc
-  // 4.2 to pessimize LexTokenInternal, a very performance critical
-  // routine. Keeping as separate members with casts until a more
-  // beautiful fix presents itself.
+  // Conceptually these next two fields could be in a union.  However, this
+  // causes gcc 4.2 to pessimize LexTokenInternal, a very performance critical
+  // routine. Keeping as separate members with casts until a more beautiful fix
+  // presents itself.
 
   /// UintData - This holds either the length of the token text, when
   /// a normal token, or the end of the SourceRange when an annotation
   /// token.
   unsigned UintData;
 
-  /// PtrData - For normal tokens, this points to the uniqued
-  /// information for the identifier (if an identifier token) or
-  /// null. For annotation tokens, this points to information specific
-  /// to the annotation token.
+  /// PtrData - This is a union of four different pointer types, which depends
+  /// on what type of token this is:
+  ///  Identifiers, keywords, etc:
+  ///    This is an IdentifierInfo*, which contains the uniqued identifier
+  ///    spelling.
+  ///  Literals:  isLiteral() returns true.
+  ///    This is a pointer to the start of the token in a text buffer, which
+  ///    may be dirty (have trigraphs / escaped newlines).
+  ///  Annotations (resolved type names, C++ scopes, etc): isAnnotation().
+  ///    This is a pointer to sema-specific data for the annotation token.
+  ///  Other:
+  ///    This is null.
   void *PtrData;
 
   /// Kind - The actual flavor of token this is.
@@ -77,32 +84,40 @@ public:
   bool is(tok::TokenKind K) const { return Kind == (unsigned) K; }
   bool isNot(tok::TokenKind K) const { return Kind != (unsigned) K; }
 
-  bool isAnnotationToken() const { 
+  /// isLiteral - Return true if this is a "literal", like a numeric
+  /// constant, string, etc.
+  bool isLiteral() const {
+    return is(tok::numeric_constant) || is(tok::char_constant) ||
+           is(tok::string_literal) || is(tok::wide_string_literal) ||
+           is(tok::angle_string_literal);
+  }
+
+  bool isAnnotation() const { 
     return is(tok::annot_typename) || 
            is(tok::annot_cxxscope) ||
            is(tok::annot_template_id);
   }
-
+  
   /// getLocation - Return a source location identifier for the specified
   /// offset in the current file.
   SourceLocation getLocation() const { return Loc; }
   unsigned getLength() const {
-    assert(!isAnnotationToken() && "Annotation tokens have no length field");
+    assert(!isAnnotation() && "Annotation tokens have no length field");
     return UintData;
   }
 
   void setLocation(SourceLocation L) { Loc = L; }
   void setLength(unsigned Len) {
-    assert(!isAnnotationToken() && "Annotation tokens have no length field");
+    assert(!isAnnotation() && "Annotation tokens have no length field");
     UintData = Len;
   }
 
   SourceLocation getAnnotationEndLoc() const {
-    assert(isAnnotationToken() && "Used AnnotEndLocID on non-annotation token");
+    assert(isAnnotation() && "Used AnnotEndLocID on non-annotation token");
     return SourceLocation::getFromRawEncoding(UintData);
   }
   void setAnnotationEndLoc(SourceLocation L) {
-    assert(isAnnotationToken() && "Used AnnotEndLocID on non-annotation token");
+    assert(isAnnotation() && "Used AnnotEndLocID on non-annotation token");
     UintData = L.getRawEncoding();
   }
 
@@ -130,19 +145,32 @@ public:
   }
   
   IdentifierInfo *getIdentifierInfo() const {
-    assert(!isAnnotationToken() && "Used IdentInfo on annotation token");
+    assert(!isAnnotation() && "Used IdentInfo on annotation token!");
+    if (isLiteral()) return 0;
     return (IdentifierInfo*) PtrData;
   }
   void setIdentifierInfo(IdentifierInfo *II) {
     PtrData = (void*) II;
   }
-
+  
+  /// getLiteralData - For a literal token (numeric constant, string, etc), this
+  /// returns a pointer to the start of it in the text buffer if known, null
+  /// otherwise.
+  const char *getLiteralData() const {
+    assert(isLiteral() && "Cannot get literal data of non-literal");
+    return reinterpret_cast<const char*>(PtrData);
+  }
+  void setLiteralData(const char *Ptr) {
+    assert(isLiteral() && "Cannot set literal data of non-literal");
+    PtrData = (void*)Ptr;
+  }
+  
   void *getAnnotationValue() const {
-    assert(isAnnotationToken() && "Used AnnotVal on non-annotation token");
+    assert(isAnnotation() && "Used AnnotVal on non-annotation token");
     return PtrData;
   }
   void setAnnotationValue(void *val) {
-    assert(isAnnotationToken() && "Used AnnotVal on non-annotation token");
+    assert(isAnnotation() && "Used AnnotVal on non-annotation token");
     PtrData = val;
   }
   

include/clang/Parse/Parser.h

@@ -690,10 +690,11 @@ private:
     if (!Tok.is(tok::identifier))
       return false;
     
-    if (Actions.isTypeName(*Tok.getIdentifierInfo(), CurScope))
+    IdentifierInfo *II = Tok.getIdentifierInfo();
+    if (Actions.isTypeName(*II, CurScope))
       return true;
     
-    return Tok.getIdentifierInfo() == Ident_super;
+    return II == Ident_super;
   }
 
   OwningExprResult ParseObjCAtExpression(SourceLocation AtLocation);

lib/Lex/Lexer.cpp

@@ -624,7 +624,9 @@ void Lexer::LexNumericConstant(Token &Result, const char *CurPtr) {
     return LexNumericConstant(Result, ConsumeChar(CurPtr, Size, Result));
   
   // Update the location of token as well as BufferPtr.
+  const char *TokStart = BufferPtr;
   FormTokenWithChars(Result, CurPtr, tok::numeric_constant);
+  Result.setLiteralData(TokStart);
 }
 
 /// LexStringLiteral - Lex the remainder of a string literal, after having lexed
@@ -655,8 +657,10 @@ void Lexer::LexStringLiteral(Token &Result, const char *CurPtr, bool Wide) {
     Diag(NulCharacter, diag::null_in_string);
 
   // Update the location of the token as well as the BufferPtr instance var.
+  const char *TokStart = BufferPtr;
   FormTokenWithChars(Result, CurPtr,
                      Wide ? tok::wide_string_literal : tok::string_literal);
+  Result.setLiteralData(TokStart);
 }
 
 /// LexAngledStringLiteral - Lex the remainder of an angled string literal,
@@ -687,7 +691,9 @@ void Lexer::LexAngledStringLiteral(Token &Result, const char *CurPtr) {
     Diag(NulCharacter, diag::null_in_string);
   
   // Update the location of token as well as BufferPtr.
+  const char *TokStart = BufferPtr;
   FormTokenWithChars(Result, CurPtr, tok::angle_string_literal);
+  Result.setLiteralData(TokStart);
 }
 
 
@@ -735,7 +741,9 @@ void Lexer::LexCharConstant(Token &Result, const char *CurPtr) {
     Diag(NulCharacter, diag::null_in_char);
 
   // Update the location of token as well as BufferPtr.
+  const char *TokStart = BufferPtr;
   FormTokenWithChars(Result, CurPtr, tok::char_constant);
+  Result.setLiteralData(TokStart);
 }
 
 /// SkipWhitespace - Efficiently skip over a series of whitespace characters.
@@ -901,9 +909,8 @@ bool Lexer::SaveBCPLComment(Token &Result, const char *CurPtr) {
   Spelling += "*/";    // add suffix.
   
   Result.setKind(tok::comment);
-  Result.setLocation(PP->CreateString(&Spelling[0], Spelling.size(),
-                                      Result.getLocation()));
-  Result.setLength(Spelling.size());
+  PP->CreateString(&Spelling[0], Spelling.size(), Result,
+                   Result.getLocation());
   return true;
 }
 

lib/Lex/MacroArgs.cpp

@@ -225,8 +225,7 @@ Token MacroArgs::StringifyArgument(const Token *ArgToks,
     }
   }
   
-  Tok.setLength(Result.size());
-  Tok.setLocation(PP.CreateString(&Result[0], Result.size()));
+  PP.CreateString(&Result[0], Result.size(), Tok);
   return Tok;
 }
 

lib/Lex/PPCaching.cpp

@@ -89,7 +89,7 @@ const Token &Preprocessor::PeekAhead(unsigned N) {
 }
 
 void Preprocessor::AnnotatePreviousCachedTokens(const Token &Tok) {
-  assert(Tok.isAnnotationToken() && "Expected annotation token");
+  assert(Tok.isAnnotation() && "Expected annotation token");
   assert(CachedLexPos != 0 && "Expected to have some cached tokens");
   assert(CachedTokens[CachedLexPos-1].getLocation() == Tok.getAnnotationEndLoc()
          && "The annotation should be until the most recent cached token");

lib/Lex/PPMacroExpansion.cpp

@@ -429,10 +429,15 @@ static void ComputeDATE_TIME(SourceLocation &DATELoc, SourceLocation &TIMELoc,
   char TmpBuffer[100];
   sprintf(TmpBuffer, "\"%s %2d %4d\"", Months[TM->tm_mon], TM->tm_mday, 
           TM->tm_year+1900);
-  DATELoc = PP.CreateString(TmpBuffer, strlen(TmpBuffer));
+  
+  Token TmpTok;
+  TmpTok.startToken();
+  PP.CreateString(TmpBuffer, strlen(TmpBuffer), TmpTok);
+  DATELoc = TmpTok.getLocation();
 
   sprintf(TmpBuffer, "\"%02d:%02d:%02d\"", TM->tm_hour, TM->tm_min, TM->tm_sec);
-  TIMELoc = PP.CreateString(TmpBuffer, strlen(TmpBuffer));
+  PP.CreateString(TmpBuffer, strlen(TmpBuffer), TmpTok);
+  TIMELoc = TmpTok.getLocation();
 }
 
 /// ExpandBuiltinMacro - If an identifier token is read that is to be expanded
@@ -463,8 +468,8 @@ void Preprocessor::ExpandBuiltinMacro(Token &Tok) {
             SourceMgr.getInstantiationLineNumber(Tok.getLocation()));
     unsigned Length = strlen(TmpBuffer)-1;
     Tok.setKind(tok::numeric_constant);
-    Tok.setLength(Length);
-    Tok.setLocation(CreateString(TmpBuffer, Length+1, Tok.getLocation()));
+    CreateString(TmpBuffer, Length+1, Tok, Tok.getLocation());
+    Tok.setLength(Length);  // Trim off space.
   } else if (II == Ident__FILE__ || II == Ident__BASE_FILE__) {
     SourceLocation Loc = Tok.getLocation();
     if (II == Ident__BASE_FILE__) {
@@ -480,8 +485,7 @@ void Preprocessor::ExpandBuiltinMacro(Token &Tok) {
     std::string FN =SourceMgr.getSourceName(SourceMgr.getInstantiationLoc(Loc));
     FN = '"' + Lexer::Stringify(FN) + '"';
     Tok.setKind(tok::string_literal);
-    Tok.setLength(FN.size());
-    Tok.setLocation(CreateString(&FN[0], FN.size(), Tok.getLocation()));
+    CreateString(&FN[0], FN.size(), Tok, Tok.getLocation());
   } else if (II == Ident__DATE__) {
     if (!DATELoc.isValid())
       ComputeDATE_TIME(DATELoc, TIMELoc, *this);
@@ -511,8 +515,8 @@ void Preprocessor::ExpandBuiltinMacro(Token &Tok) {
     sprintf(TmpBuffer, "%u ", Depth);
     unsigned Length = strlen(TmpBuffer)-1;
     Tok.setKind(tok::numeric_constant);
-    Tok.setLength(Length);
-    Tok.setLocation(CreateString(TmpBuffer, Length, Tok.getLocation()));
+    CreateString(TmpBuffer, Length, Tok, Tok.getLocation());
+    Tok.setLength(Length);  // Trim off space.
   } else if (II == Ident__TIMESTAMP__) {
     // MSVC, ICC, GCC, VisualAge C++ extension.  The generated string should be
     // of the form "Ddd Mmm dd hh::mm::ss yyyy", which is returned by asctime.
@@ -540,8 +544,8 @@ void Preprocessor::ExpandBuiltinMacro(Token &Tok) {
     unsigned Len = strlen(TmpBuffer);
     TmpBuffer[Len-1] = '"';  // Replace the newline with a quote.
     Tok.setKind(tok::string_literal);
-    Tok.setLength(Len);
-    Tok.setLocation(CreateString(TmpBuffer, Len+1, Tok.getLocation()));
+    CreateString(TmpBuffer, Len+1, Tok, Tok.getLocation());
+    Tok.setLength(Len);  // Trim off space.
   } else {
     assert(0 && "Unknown identifier!");
   }

lib/Lex/Pragma.cpp

@@ -156,7 +156,10 @@ void Preprocessor::Handle_Pragma(Token &Tok) {
   
   // Plop the string (including the newline and trailing null) into a buffer
   // where we can lex it.
-  SourceLocation TokLoc = CreateString(&StrVal[0], StrVal.size());
+  Token TmpTok;
+  TmpTok.startToken();
+  CreateString(&StrVal[0], StrVal.size(), TmpTok);
+  SourceLocation TokLoc = TmpTok.getLocation();
 
   // Make and enter a lexer object so that we lex and expand the tokens just
   // like any others.

lib/Lex/Preprocessor.cpp

@@ -266,13 +266,20 @@ unsigned Preprocessor::getSpelling(const Token &Tok,
   }
 
   // Otherwise, compute the start of the token in the input lexer buffer.
-  const char *TokStart = SourceMgr.getCharacterData(Tok.getLocation());
+  const char *TokStart = 0;
+  
+  if (Tok.isLiteral())
+    TokStart = Tok.getLiteralData();
+  
+  if (TokStart == 0)
+    TokStart = SourceMgr.getCharacterData(Tok.getLocation());
 
   // If this token contains nothing interesting, return it directly.
   if (!Tok.needsCleaning()) {
     Buffer = TokStart;
     return Tok.getLength();
   }
+  
   // Otherwise, hard case, relex the characters into the string.
   char *OutBuf = const_cast<char*>(Buffer);
   for (const char *Ptr = TokStart, *End = TokStart+Tok.getLength();
@@ -291,11 +298,20 @@ unsigned Preprocessor::getSpelling(const Token &Tok,
 /// CreateString - Plop the specified string into a scratch buffer and return a
 /// location for it.  If specified, the source location provides a source
 /// location for the token.
-SourceLocation Preprocessor::
-CreateString(const char *Buf, unsigned Len, SourceLocation SLoc) {
-  if (SLoc.isValid())
-    return ScratchBuf->getToken(Buf, Len, SLoc);
-  return ScratchBuf->getToken(Buf, Len);
+void Preprocessor::CreateString(const char *Buf, unsigned Len, Token &Tok,
+                                SourceLocation InstantiationLoc) {
+  Tok.setLength(Len);
+  
+  const char *DestPtr;
+  SourceLocation Loc = ScratchBuf->getToken(Buf, Len, DestPtr);
+  
+  if (InstantiationLoc.isValid())
+    Loc = SourceMgr.createInstantiationLoc(Loc, InstantiationLoc, Len);
+  Tok.setLocation(Loc);
+  
+  // If this is a literal token, set the pointer data.
+  if (Tok.isLiteral())
+    Tok.setLiteralData(DestPtr);
 }
 
 

lib/Lex/ScratchBuffer.cpp

@@ -30,10 +30,14 @@ ScratchBuffer::ScratchBuffer(SourceManager &SM) : SourceMgr(SM), CurBuffer(0) {
 /// return a SourceLocation that refers to the token.  This is just like the
 /// method below, but returns a location that indicates the physloc of the
 /// token.
-SourceLocation ScratchBuffer::getToken(const char *Buf, unsigned Len) {
+SourceLocation ScratchBuffer::getToken(const char *Buf, unsigned Len,
+                                       const char *&DestPtr) {
   if (BytesUsed+Len > ScratchBufSize)
     AllocScratchBuffer(Len);
   
+  // Return a pointer to the character data.
+  DestPtr = CurBuffer+BytesUsed;
+  
   // Copy the token data into the buffer.
   memcpy(CurBuffer+BytesUsed, Buf, Len);
 
@@ -43,16 +47,6 @@ SourceLocation ScratchBuffer::getToken(const char *Buf, unsigned Len) {
   return BufferStartLoc.getFileLocWithOffset(BytesUsed-Len);
 }
 
-
-/// getToken - Splat the specified text into a temporary MemoryBuffer and
-/// return a SourceLocation that refers to the token.  The SourceLoc value
-/// gives a virtual location that the token will appear to be from.
-SourceLocation ScratchBuffer::getToken(const char *Buf, unsigned Len,
-                                       SourceLocation SourceLoc) {
-  // Map the physloc to the specified sourceloc.
-  return SourceMgr.createInstantiationLoc(getToken(Buf, Len), SourceLoc, Len);
-}
-
 void ScratchBuffer::AllocScratchBuffer(unsigned RequestLen) {
   // Only pay attention to the requested length if it is larger than our default
   // page size.  If it is, we allocate an entire chunk for it.  This is to

lib/Lex/TokenLexer.cpp

@@ -352,6 +352,7 @@ void TokenLexer::Lex(Token &Tok) {
 /// If this returns true, the caller should immediately return the token.
 bool TokenLexer::PasteTokens(Token &Tok) {
   llvm::SmallVector<char, 128> Buffer;
+  const char *ResultTokStrPtr = 0;
   do {
     // Consume the ## operator.
     SourceLocation PasteOpLoc = Tokens[CurToken].getLocation();
@@ -386,8 +387,16 @@ bool TokenLexer::PasteTokens(Token &Tok) {
     
     // Plop the pasted result (including the trailing newline and null) into a
     // scratch buffer where we can lex it.
-    SourceLocation ResultTokLoc = PP.CreateString(&Buffer[0], Buffer.size());
+    Token ResultTokTmp;
+    ResultTokTmp.startToken();
     
+    // Claim that the tmp token is a string_literal so that we can get the
+    // character pointer back from CreateString.
+    ResultTokTmp.setKind(tok::string_literal);
+    PP.CreateString(&Buffer[0], Buffer.size(), ResultTokTmp);
+    SourceLocation ResultTokLoc = ResultTokTmp.getLocation();
+    ResultTokStrPtr = ResultTokTmp.getLiteralData();
+
     // Lex the resultant pasted token into Result.
     Token Result;
     
@@ -405,20 +414,16 @@ bool TokenLexer::PasteTokens(Token &Tok) {
       assert(ResultTokLoc.isFileID() &&
              "Should be a raw location into scratch buffer");
       SourceManager &SourceMgr = PP.getSourceManager();
-      std::pair<FileID, unsigned> LocInfo =
-        SourceMgr.getDecomposedLoc(ResultTokLoc);
+      FileID LocFileID = SourceMgr.getFileID(ResultTokLoc);
       
-      const char *ScratchBufStart =SourceMgr.getBufferData(LocInfo.first).first;
+      const char *ScratchBufStart = SourceMgr.getBufferData(LocFileID).first;
       
       // Make a lexer to lex this string from.  Lex just this one token.
-      const char *ResultStrData = ScratchBufStart+LocInfo.second;
-      
       // Make a lexer object so that we lex and expand the paste result.
-      Lexer TL(SourceMgr.getLocForStartOfFile(LocInfo.first),
-               PP.getLangOptions(), 
-               ScratchBufStart,
-               ResultStrData, 
-               ResultStrData+LHSLen+RHSLen /*don't include null*/);
+      Lexer TL(SourceMgr.getLocForStartOfFile(LocFileID),
+               PP.getLangOptions(), ScratchBufStart,
+               ResultTokStrPtr, 
+               ResultTokStrPtr+LHSLen+RHSLen /*don't include null*/);
       
       // Lex a token in raw mode.  This way it won't look up identifiers
       // automatically, lexing off the end will return an eof token, and
@@ -442,12 +447,12 @@ bool TokenLexer::PasteTokens(Token &Tok) {
           RHS.is(tok::slash)) {
         HandleMicrosoftCommentPaste(Tok);
         return true;
-      } else {
-        // TODO: If not in assembler language mode.
-        PP.Diag(PasteOpLoc, diag::err_pp_bad_paste)
-          << std::string(Buffer.begin(), Buffer.end()-1);
-        return false;
       }
+      
+      // TODO: If not in assembler language mode.
+      PP.Diag(PasteOpLoc, diag::err_pp_bad_paste)
+        << std::string(Buffer.begin(), Buffer.end()-1);
+      return false;
     }
     
     // Turn ## into 'unknown' to avoid # ## # from looking like a paste
@@ -471,7 +476,7 @@ bool TokenLexer::PasteTokens(Token &Tok) {
   if (Tok.is(tok::identifier)) {
     // Look up the identifier info for the token.  We disabled identifier lookup
     // by saying we're skipping contents, so we need to do this manually.
-    Tok.setIdentifierInfo(PP.LookUpIdentifierInfo(Tok));
+    Tok.setIdentifierInfo(PP.LookUpIdentifierInfo(Tok, ResultTokStrPtr));
   }
   return false;
 }

lib/Rewrite/TokenRewriter.cpp

@@ -78,14 +78,15 @@ TokenRewriter::AddToken(const Token &T, TokenRefTy Where) {
   
 
 TokenRewriter::token_iterator
-TokenRewriter::AddTokenBefore(token_iterator I, const char *Val){
+TokenRewriter::AddTokenBefore(token_iterator I, const char *Val) {
   unsigned Len = strlen(Val);
   
   // Plop the string into the scratch buffer, then create a token for this
   // string.
   Token Tok;
   Tok.startToken();
-  Tok.setLocation(ScratchBuf->getToken(Val, Len));
+  const char *Spelling;
+  Tok.setLocation(ScratchBuf->getToken(Val, Len, Spelling));
   Tok.setLength(Len);
   
   // TODO: Form a whole lexer around this and relex the token!  For now, just