1 //===--- TokenConcatenation.cpp - Token Concatenation Avoidance -----------===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This file implements the TokenConcatenation class. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #include "clang/Lex/TokenConcatenation.h" 15 #include "clang/Lex/Preprocessor.h" 16 #include "llvm/Support/ErrorHandling.h" 17 using namespace clang; 18 19 20 /// IsStringPrefix - Return true if Str is a string prefix. 21 /// 'L', 'u', 'U', or 'u8'. Including raw versions. 22 static bool IsStringPrefix(StringRef Str, bool CPlusPlus0x) { 23 24 if (Str[0] == 'L' || 25 (CPlusPlus0x && (Str[0] == 'u' || Str[0] == 'U' || Str[0] == 'R'))) { 26 27 if (Str.size() == 1) 28 return true; // "L", "u", "U", and "R" 29 30 // Check for raw flavors. Need to make sure the first character wasn't 31 // already R. Need CPlusPlus0x check for "LR". 32 if (Str[1] == 'R' && Str[0] != 'R' && Str.size() == 2 && CPlusPlus0x) 33 return true; // "LR", "uR", "UR" 34 35 // Check for "u8" and "u8R" 36 if (Str[0] == 'u' && Str[1] == '8') { 37 if (Str.size() == 2) return true; // "u8" 38 if (Str.size() == 3 && Str[2] == 'R') return true; // "u8R" 39 } 40 } 41 42 return false; 43 } 44 45 /// IsIdentifierStringPrefix - Return true if the spelling of the token 46 /// is literally 'L', 'u', 'U', or 'u8'. Including raw versions. 47 bool TokenConcatenation::IsIdentifierStringPrefix(const Token &Tok) const { 48 const LangOptions &LangOpts = PP.getLangOptions(); 49 50 if (!Tok.needsCleaning()) { 51 if (Tok.getLength() < 1 || Tok.getLength() > 3) 52 return false; 53 SourceManager &SM = PP.getSourceManager(); 54 const char *Ptr = SM.getCharacterData(SM.getSpellingLoc(Tok.getLocation())); 55 return IsStringPrefix(StringRef(Ptr, Tok.getLength()), 56 LangOpts.CPlusPlus0x); 57 } 58 59 if (Tok.getLength() < 256) { 60 char Buffer[256]; 61 const char *TokPtr = Buffer; 62 unsigned length = PP.getSpelling(Tok, TokPtr); 63 return IsStringPrefix(StringRef(TokPtr, length), LangOpts.CPlusPlus0x); 64 } 65 66 return IsStringPrefix(StringRef(PP.getSpelling(Tok)), LangOpts.CPlusPlus0x); 67 } 68 69 TokenConcatenation::TokenConcatenation(Preprocessor &pp) : PP(pp) { 70 memset(TokenInfo, 0, sizeof(TokenInfo)); 71 72 // These tokens have custom code in AvoidConcat. 73 TokenInfo[tok::identifier ] |= aci_custom; 74 TokenInfo[tok::numeric_constant] |= aci_custom_firstchar; 75 TokenInfo[tok::period ] |= aci_custom_firstchar; 76 TokenInfo[tok::amp ] |= aci_custom_firstchar; 77 TokenInfo[tok::plus ] |= aci_custom_firstchar; 78 TokenInfo[tok::minus ] |= aci_custom_firstchar; 79 TokenInfo[tok::slash ] |= aci_custom_firstchar; 80 TokenInfo[tok::less ] |= aci_custom_firstchar; 81 TokenInfo[tok::greater ] |= aci_custom_firstchar; 82 TokenInfo[tok::pipe ] |= aci_custom_firstchar; 83 TokenInfo[tok::percent ] |= aci_custom_firstchar; 84 TokenInfo[tok::colon ] |= aci_custom_firstchar; 85 TokenInfo[tok::hash ] |= aci_custom_firstchar; 86 TokenInfo[tok::arrow ] |= aci_custom_firstchar; 87 88 // These tokens change behavior if followed by an '='. 89 TokenInfo[tok::amp ] |= aci_avoid_equal; // &= 90 TokenInfo[tok::plus ] |= aci_avoid_equal; // += 91 TokenInfo[tok::minus ] |= aci_avoid_equal; // -= 92 TokenInfo[tok::slash ] |= aci_avoid_equal; // /= 93 TokenInfo[tok::less ] |= aci_avoid_equal; // <= 94 TokenInfo[tok::greater ] |= aci_avoid_equal; // >= 95 TokenInfo[tok::pipe ] |= aci_avoid_equal; // |= 96 TokenInfo[tok::percent ] |= aci_avoid_equal; // %= 97 TokenInfo[tok::star ] |= aci_avoid_equal; // *= 98 TokenInfo[tok::exclaim ] |= aci_avoid_equal; // != 99 TokenInfo[tok::lessless ] |= aci_avoid_equal; // <<= 100 TokenInfo[tok::greatergreater] |= aci_avoid_equal; // >>= 101 TokenInfo[tok::caret ] |= aci_avoid_equal; // ^= 102 TokenInfo[tok::equal ] |= aci_avoid_equal; // == 103 } 104 105 /// GetFirstChar - Get the first character of the token \arg Tok, 106 /// avoiding calls to getSpelling where possible. 107 static char GetFirstChar(Preprocessor &PP, const Token &Tok) { 108 if (IdentifierInfo *II = Tok.getIdentifierInfo()) { 109 // Avoid spelling identifiers, the most common form of token. 110 return II->getNameStart()[0]; 111 } else if (!Tok.needsCleaning()) { 112 if (Tok.isLiteral() && Tok.getLiteralData()) { 113 return *Tok.getLiteralData(); 114 } else { 115 SourceManager &SM = PP.getSourceManager(); 116 return *SM.getCharacterData(SM.getSpellingLoc(Tok.getLocation())); 117 } 118 } else if (Tok.getLength() < 256) { 119 char Buffer[256]; 120 const char *TokPtr = Buffer; 121 PP.getSpelling(Tok, TokPtr); 122 return TokPtr[0]; 123 } else { 124 return PP.getSpelling(Tok)[0]; 125 } 126 } 127 128 /// AvoidConcat - If printing PrevTok immediately followed by Tok would cause 129 /// the two individual tokens to be lexed as a single token, return true 130 /// (which causes a space to be printed between them). This allows the output 131 /// of -E mode to be lexed to the same token stream as lexing the input 132 /// directly would. 133 /// 134 /// This code must conservatively return true if it doesn't want to be 100% 135 /// accurate. This will cause the output to include extra space characters, 136 /// but the resulting output won't have incorrect concatenations going on. 137 /// Examples include "..", which we print with a space between, because we 138 /// don't want to track enough to tell "x.." from "...". 139 bool TokenConcatenation::AvoidConcat(const Token &PrevPrevTok, 140 const Token &PrevTok, 141 const Token &Tok) const { 142 // First, check to see if the tokens were directly adjacent in the original 143 // source. If they were, it must be okay to stick them together: if there 144 // were an issue, the tokens would have been lexed differently. 145 if (PrevTok.getLocation().isFileID() && Tok.getLocation().isFileID() && 146 PrevTok.getLocation().getLocWithOffset(PrevTok.getLength()) == 147 Tok.getLocation()) 148 return false; 149 150 tok::TokenKind PrevKind = PrevTok.getKind(); 151 if (PrevTok.getIdentifierInfo()) // Language keyword or named operator. 152 PrevKind = tok::identifier; 153 154 // Look up information on when we should avoid concatenation with prevtok. 155 unsigned ConcatInfo = TokenInfo[PrevKind]; 156 157 // If prevtok never causes a problem for anything after it, return quickly. 158 if (ConcatInfo == 0) return false; 159 160 if (ConcatInfo & aci_avoid_equal) { 161 // If the next token is '=' or '==', avoid concatenation. 162 if (Tok.is(tok::equal) || Tok.is(tok::equalequal)) 163 return true; 164 ConcatInfo &= ~aci_avoid_equal; 165 } 166 167 if (ConcatInfo == 0) return false; 168 169 // Basic algorithm: we look at the first character of the second token, and 170 // determine whether it, if appended to the first token, would form (or 171 // would contribute) to a larger token if concatenated. 172 char FirstChar = 0; 173 if (ConcatInfo & aci_custom) { 174 // If the token does not need to know the first character, don't get it. 175 } else { 176 FirstChar = GetFirstChar(PP, Tok); 177 } 178 179 switch (PrevKind) { 180 default: 181 llvm_unreachable("InitAvoidConcatTokenInfo built wrong"); 182 183 case tok::raw_identifier: 184 llvm_unreachable("tok::raw_identifier in non-raw lexing mode!"); 185 186 case tok::identifier: // id+id or id+number or id+L"foo". 187 // id+'.'... will not append. 188 if (Tok.is(tok::numeric_constant)) 189 return GetFirstChar(PP, Tok) != '.'; 190 191 if (Tok.getIdentifierInfo() || Tok.is(tok::wide_string_literal) || 192 Tok.is(tok::utf8_string_literal) || Tok.is(tok::utf16_string_literal) || 193 Tok.is(tok::utf32_string_literal) || Tok.is(tok::wide_char_constant) || 194 Tok.is(tok::utf16_char_constant) || Tok.is(tok::utf32_char_constant)) 195 return true; 196 197 // If this isn't identifier + string, we're done. 198 if (Tok.isNot(tok::char_constant) && Tok.isNot(tok::string_literal)) 199 return false; 200 201 // Otherwise, this is a narrow character or string. If the *identifier* 202 // is a literal 'L', 'u8', 'u' or 'U', avoid pasting L "foo" -> L"foo". 203 return IsIdentifierStringPrefix(PrevTok); 204 case tok::numeric_constant: 205 return isalnum(FirstChar) || Tok.is(tok::numeric_constant) || 206 FirstChar == '+' || FirstChar == '-' || FirstChar == '.'; 207 case tok::period: // ..., .*, .1234 208 return (FirstChar == '.' && PrevPrevTok.is(tok::period)) || 209 isdigit(FirstChar) || 210 (PP.getLangOptions().CPlusPlus && FirstChar == '*'); 211 case tok::amp: // && 212 return FirstChar == '&'; 213 case tok::plus: // ++ 214 return FirstChar == '+'; 215 case tok::minus: // --, ->, ->* 216 return FirstChar == '-' || FirstChar == '>'; 217 case tok::slash: //, /*, // 218 return FirstChar == '*' || FirstChar == '/'; 219 case tok::less: // <<, <<=, <:, <% 220 return FirstChar == '<' || FirstChar == ':' || FirstChar == '%'; 221 case tok::greater: // >>, >>= 222 return FirstChar == '>'; 223 case tok::pipe: // || 224 return FirstChar == '|'; 225 case tok::percent: // %>, %: 226 return FirstChar == '>' || FirstChar == ':'; 227 case tok::colon: // ::, :> 228 return FirstChar == '>' || 229 (PP.getLangOptions().CPlusPlus && FirstChar == ':'); 230 case tok::hash: // ##, #@, %:%: 231 return FirstChar == '#' || FirstChar == '@' || FirstChar == '%'; 232 case tok::arrow: // ->* 233 return PP.getLangOptions().CPlusPlus && FirstChar == '*'; 234 } 235 } 236