lib/Lex/TokenConcatenation.cpp

//===--- TokenConcatenation.cpp - Token Concatenation Avoidance -----------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file implements the TokenConcatenation class.
//
//===----------------------------------------------------------------------===//

#include "clang/Lex/TokenConcatenation.h"
#include "clang/Lex/Preprocessor.h"
#include "llvm/Support/ErrorHandling.h"
using namespace clang;


/// IsIdentifierStringPrefix - Return true if the spelling of the token
/// is literally 'L', 'u', 'U', or 'u8'.
bool TokenConcatenation::IsIdentifierStringPrefix(const Token &Tok) const {
  const LangOptions &LangOpts = PP.getLangOptions();

  if (!Tok.needsCleaning()) {
    if (Tok.getLength() != 1 && Tok.getLength() != 2)
      return false;
    SourceManager &SM = PP.getSourceManager();
    const char *Ptr = SM.getCharacterData(SM.getSpellingLoc(Tok.getLocation()));
    if (Tok.getLength() == 1)
      return Ptr[0] == 'L' ||
             (LangOpts.CPlusPlus0x && (Ptr[0] == 'u' || Ptr[0] == 'U'));
    if (Tok.getLength() == 2)
      return LangOpts.CPlusPlus0x && Ptr[0] == 'u' && Ptr[1] == '8';
  }

  if (Tok.getLength() < 256) {
    char Buffer[256];
    const char *TokPtr = Buffer;
    unsigned length = PP.getSpelling(Tok, TokPtr);
    if (length == 1)
      return TokPtr[0] == 'L' ||
             (LangOpts.CPlusPlus0x && (TokPtr[0] == 'u' || TokPtr[0] == 'U'));
    if (length == 2)
      return LangOpts.CPlusPlus0x && TokPtr[0] == 'u' && TokPtr[1] == '8';
    return false;
  }

  std::string TokStr = PP.getSpelling(Tok);
  return TokStr == "L" || (LangOpts.CPlusPlus0x && (TokStr == "u8" ||
                                                    TokStr == "u" ||
                                                    TokStr == "U"));
}

TokenConcatenation::TokenConcatenation(Preprocessor &pp) : PP(pp) {
  memset(TokenInfo, 0, sizeof(TokenInfo));

  // These tokens have custom code in AvoidConcat.
  TokenInfo[tok::identifier      ] |= aci_custom;
  TokenInfo[tok::numeric_constant] |= aci_custom_firstchar;
  TokenInfo[tok::period          ] |= aci_custom_firstchar;
  TokenInfo[tok::amp             ] |= aci_custom_firstchar;
  TokenInfo[tok::plus            ] |= aci_custom_firstchar;
  TokenInfo[tok::minus           ] |= aci_custom_firstchar;
  TokenInfo[tok::slash           ] |= aci_custom_firstchar;
  TokenInfo[tok::less            ] |= aci_custom_firstchar;
  TokenInfo[tok::greater         ] |= aci_custom_firstchar;
  TokenInfo[tok::pipe            ] |= aci_custom_firstchar;
  TokenInfo[tok::percent         ] |= aci_custom_firstchar;
  TokenInfo[tok::colon           ] |= aci_custom_firstchar;
  TokenInfo[tok::hash            ] |= aci_custom_firstchar;
  TokenInfo[tok::arrow           ] |= aci_custom_firstchar;

  // These tokens change behavior if followed by an '='.
  TokenInfo[tok::amp         ] |= aci_avoid_equal;           // &=
  TokenInfo[tok::plus        ] |= aci_avoid_equal;           // +=
  TokenInfo[tok::minus       ] |= aci_avoid_equal;           // -=
  TokenInfo[tok::slash       ] |= aci_avoid_equal;           // /=
  TokenInfo[tok::less        ] |= aci_avoid_equal;           // <=
  TokenInfo[tok::greater     ] |= aci_avoid_equal;           // >=
  TokenInfo[tok::pipe        ] |= aci_avoid_equal;           // |=
  TokenInfo[tok::percent     ] |= aci_avoid_equal;           // %=
  TokenInfo[tok::star        ] |= aci_avoid_equal;           // *=
  TokenInfo[tok::exclaim     ] |= aci_avoid_equal;           // !=
  TokenInfo[tok::lessless    ] |= aci_avoid_equal;           // <<=
  TokenInfo[tok::greatergreater] |= aci_avoid_equal;         // >>=
  TokenInfo[tok::caret       ] |= aci_avoid_equal;           // ^=
  TokenInfo[tok::equal       ] |= aci_avoid_equal;           // ==
}

/// GetFirstChar - Get the first character of the token \arg Tok,
/// avoiding calls to getSpelling where possible.
static char GetFirstChar(Preprocessor &PP, const Token &Tok) {
  if (IdentifierInfo *II = Tok.getIdentifierInfo()) {
    // Avoid spelling identifiers, the most common form of token.
    return II->getNameStart()[0];
  } else if (!Tok.needsCleaning()) {
    if (Tok.isLiteral() && Tok.getLiteralData()) {
      return *Tok.getLiteralData();
    } else {
      SourceManager &SM = PP.getSourceManager();
      return *SM.getCharacterData(SM.getSpellingLoc(Tok.getLocation()));
    }
  } else if (Tok.getLength() < 256) {
    char Buffer[256];
    const char *TokPtr = Buffer;
    PP.getSpelling(Tok, TokPtr);
    return TokPtr[0];
  } else {
    return PP.getSpelling(Tok)[0];
  }
}

/// AvoidConcat - If printing PrevTok immediately followed by Tok would cause
/// the two individual tokens to be lexed as a single token, return true
/// (which causes a space to be printed between them).  This allows the output
/// of -E mode to be lexed to the same token stream as lexing the input
/// directly would.
///
/// This code must conservatively return true if it doesn't want to be 100%
/// accurate.  This will cause the output to include extra space characters,
/// but the resulting output won't have incorrect concatenations going on.
/// Examples include "..", which we print with a space between, because we
/// don't want to track enough to tell "x.." from "...".
bool TokenConcatenation::AvoidConcat(const Token &PrevPrevTok,
                                     const Token &PrevTok,
                                     const Token &Tok) const {
  // First, check to see if the tokens were directly adjacent in the original
  // source.  If they were, it must be okay to stick them together: if there
  // were an issue, the tokens would have been lexed differently.
  if (PrevTok.getLocation().isFileID() && Tok.getLocation().isFileID() &&
      PrevTok.getLocation().getFileLocWithOffset(PrevTok.getLength()) ==
        Tok.getLocation())
    return false;

  tok::TokenKind PrevKind = PrevTok.getKind();
  if (PrevTok.getIdentifierInfo())  // Language keyword or named operator.
    PrevKind = tok::identifier;

  // Look up information on when we should avoid concatenation with prevtok.
  unsigned ConcatInfo = TokenInfo[PrevKind];

  // If prevtok never causes a problem for anything after it, return quickly.
  if (ConcatInfo == 0) return false;

  if (ConcatInfo & aci_avoid_equal) {
    // If the next token is '=' or '==', avoid concatenation.
    if (Tok.is(tok::equal) || Tok.is(tok::equalequal))
      return true;
    ConcatInfo &= ~aci_avoid_equal;
  }

  if (ConcatInfo == 0) return false;

  // Basic algorithm: we look at the first character of the second token, and
  // determine whether it, if appended to the first token, would form (or
  // would contribute) to a larger token if concatenated.
  char FirstChar = 0;
  if (ConcatInfo & aci_custom) {
    // If the token does not need to know the first character, don't get it.
  } else {
    FirstChar = GetFirstChar(PP, Tok);
  }

  switch (PrevKind) {
  default:
    llvm_unreachable("InitAvoidConcatTokenInfo built wrong");
    return true;

  case tok::raw_identifier:
    llvm_unreachable("tok::raw_identifier in non-raw lexing mode!");
    return true;

  case tok::identifier:   // id+id or id+number or id+L"foo".
    // id+'.'... will not append.
    if (Tok.is(tok::numeric_constant))
      return GetFirstChar(PP, Tok) != '.';

    if (Tok.getIdentifierInfo() || Tok.is(tok::wide_string_literal) ||
        Tok.is(tok::utf8_string_literal) || Tok.is(tok::utf16_string_literal) ||
        Tok.is(tok::utf32_string_literal) || Tok.is(tok::wide_char_constant) ||
        Tok.is(tok::utf16_char_constant) || Tok.is(tok::utf32_char_constant))
      return true;

    // If this isn't identifier + string, we're done.
    if (Tok.isNot(tok::char_constant) && Tok.isNot(tok::string_literal))
      return false;

    // Otherwise, this is a narrow character or string.  If the *identifier*
    // is a literal 'L', 'u8', 'u' or 'U', avoid pasting L "foo" -> L"foo".
    return IsIdentifierStringPrefix(PrevTok);
  case tok::numeric_constant:
    return isalnum(FirstChar) || Tok.is(tok::numeric_constant) ||
    FirstChar == '+' || FirstChar == '-' || FirstChar == '.';
  case tok::period:          // ..., .*, .1234
    return (FirstChar == '.' && PrevPrevTok.is(tok::period)) ||
    isdigit(FirstChar) ||
    (PP.getLangOptions().CPlusPlus && FirstChar == '*');
  case tok::amp:             // &&
    return FirstChar == '&';
  case tok::plus:            // ++
    return FirstChar == '+';
  case tok::minus:           // --, ->, ->*
    return FirstChar == '-' || FirstChar == '>';
  case tok::slash:           //, /*, //
    return FirstChar == '*' || FirstChar == '/';
  case tok::less:            // <<, <<=, <:, <%
    return FirstChar == '<' || FirstChar == ':' || FirstChar == '%';
  case tok::greater:         // >>, >>=
    return FirstChar == '>';
  case tok::pipe:            // ||
    return FirstChar == '|';
  case tok::percent:         // %>, %:
    return FirstChar == '>' || FirstChar == ':';
  case tok::colon:           // ::, :>
    return FirstChar == '>' ||
    (PP.getLangOptions().CPlusPlus && FirstChar == ':');
  case tok::hash:            // ##, #@, %:%:
    return FirstChar == '#' || FirstChar == '@' || FirstChar == '%';
  case tok::arrow:           // ->*
    return PP.getLangOptions().CPlusPlus && FirstChar == '*';
  }
}