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/Basic/CharInfo.h" 16 #include "clang/Lex/Preprocessor.h" 17 #include "llvm/Support/ErrorHandling.h" 18 using namespace clang; 19 20 21 /// IsStringPrefix - Return true if Str is a string prefix. 22 /// 'L', 'u', 'U', or 'u8'. Including raw versions. 23 static bool IsStringPrefix(StringRef Str, bool CPlusPlus11) { 24 25 if (Str[0] == 'L' || 26 (CPlusPlus11 && (Str[0] == 'u' || Str[0] == 'U' || Str[0] == 'R'))) { 27 28 if (Str.size() == 1) 29 return true; // "L", "u", "U", and "R" 30 31 // Check for raw flavors. Need to make sure the first character wasn't 32 // already R. Need CPlusPlus11 check for "LR". 33 if (Str[1] == 'R' && Str[0] != 'R' && Str.size() == 2 && CPlusPlus11) 34 return true; // "LR", "uR", "UR" 35 36 // Check for "u8" and "u8R" 37 if (Str[0] == 'u' && Str[1] == '8') { 38 if (Str.size() == 2) return true; // "u8" 39 if (Str.size() == 3 && Str[2] == 'R') return true; // "u8R" 40 } 41 } 42 43 return false; 44 } 45 46 /// IsIdentifierStringPrefix - Return true if the spelling of the token 47 /// is literally 'L', 'u', 'U', or 'u8'. Including raw versions. 48 bool TokenConcatenation::IsIdentifierStringPrefix(const Token &Tok) const { 49 const LangOptions &LangOpts = PP.getLangOpts(); 50 51 if (!Tok.needsCleaning()) { 52 if (Tok.getLength() < 1 || Tok.getLength() > 3) 53 return false; 54 SourceManager &SM = PP.getSourceManager(); 55 const char *Ptr = SM.getCharacterData(SM.getSpellingLoc(Tok.getLocation())); 56 return IsStringPrefix(StringRef(Ptr, Tok.getLength()), 57 LangOpts.CPlusPlus11); 58 } 59 60 if (Tok.getLength() < 256) { 61 char Buffer[256]; 62 const char *TokPtr = Buffer; 63 unsigned length = PP.getSpelling(Tok, TokPtr); 64 return IsStringPrefix(StringRef(TokPtr, length), LangOpts.CPlusPlus11); 65 } 66 67 return IsStringPrefix(StringRef(PP.getSpelling(Tok)), LangOpts.CPlusPlus11); 68 } 69 70 TokenConcatenation::TokenConcatenation(Preprocessor &pp) : PP(pp) { 71 memset(TokenInfo, 0, sizeof(TokenInfo)); 72 73 // These tokens have custom code in AvoidConcat. 74 TokenInfo[tok::identifier ] |= aci_custom; 75 TokenInfo[tok::numeric_constant] |= aci_custom_firstchar; 76 TokenInfo[tok::period ] |= aci_custom_firstchar; 77 TokenInfo[tok::amp ] |= aci_custom_firstchar; 78 TokenInfo[tok::plus ] |= aci_custom_firstchar; 79 TokenInfo[tok::minus ] |= aci_custom_firstchar; 80 TokenInfo[tok::slash ] |= aci_custom_firstchar; 81 TokenInfo[tok::less ] |= aci_custom_firstchar; 82 TokenInfo[tok::greater ] |= aci_custom_firstchar; 83 TokenInfo[tok::pipe ] |= aci_custom_firstchar; 84 TokenInfo[tok::percent ] |= aci_custom_firstchar; 85 TokenInfo[tok::colon ] |= aci_custom_firstchar; 86 TokenInfo[tok::hash ] |= aci_custom_firstchar; 87 TokenInfo[tok::arrow ] |= aci_custom_firstchar; 88 89 // These tokens have custom code in C++11 mode. 90 if (PP.getLangOpts().CPlusPlus11) { 91 TokenInfo[tok::string_literal ] |= aci_custom; 92 TokenInfo[tok::wide_string_literal ] |= aci_custom; 93 TokenInfo[tok::utf8_string_literal ] |= aci_custom; 94 TokenInfo[tok::utf16_string_literal] |= aci_custom; 95 TokenInfo[tok::utf32_string_literal] |= aci_custom; 96 TokenInfo[tok::char_constant ] |= aci_custom; 97 TokenInfo[tok::wide_char_constant ] |= aci_custom; 98 TokenInfo[tok::utf16_char_constant ] |= aci_custom; 99 TokenInfo[tok::utf32_char_constant ] |= aci_custom; 100 } 101 102 // These tokens have custom code in C++17 mode. 103 if (PP.getLangOpts().CPlusPlus17) 104 TokenInfo[tok::utf8_char_constant] |= aci_custom; 105 106 // These tokens have custom code in C++2a mode. 107 if (PP.getLangOpts().CPlusPlus2a) 108 TokenInfo[tok::lessequal ] |= aci_custom_firstchar; 109 110 // These tokens change behavior if followed by an '='. 111 TokenInfo[tok::amp ] |= aci_avoid_equal; // &= 112 TokenInfo[tok::plus ] |= aci_avoid_equal; // += 113 TokenInfo[tok::minus ] |= aci_avoid_equal; // -= 114 TokenInfo[tok::slash ] |= aci_avoid_equal; // /= 115 TokenInfo[tok::less ] |= aci_avoid_equal; // <= 116 TokenInfo[tok::greater ] |= aci_avoid_equal; // >= 117 TokenInfo[tok::pipe ] |= aci_avoid_equal; // |= 118 TokenInfo[tok::percent ] |= aci_avoid_equal; // %= 119 TokenInfo[tok::star ] |= aci_avoid_equal; // *= 120 TokenInfo[tok::exclaim ] |= aci_avoid_equal; // != 121 TokenInfo[tok::lessless ] |= aci_avoid_equal; // <<= 122 TokenInfo[tok::greatergreater] |= aci_avoid_equal; // >>= 123 TokenInfo[tok::caret ] |= aci_avoid_equal; // ^= 124 TokenInfo[tok::equal ] |= aci_avoid_equal; // == 125 } 126 127 /// GetFirstChar - Get the first character of the token \arg Tok, 128 /// avoiding calls to getSpelling where possible. 129 static char GetFirstChar(Preprocessor &PP, const Token &Tok) { 130 if (IdentifierInfo *II = Tok.getIdentifierInfo()) { 131 // Avoid spelling identifiers, the most common form of token. 132 return II->getNameStart()[0]; 133 } else if (!Tok.needsCleaning()) { 134 if (Tok.isLiteral() && Tok.getLiteralData()) { 135 return *Tok.getLiteralData(); 136 } else { 137 SourceManager &SM = PP.getSourceManager(); 138 return *SM.getCharacterData(SM.getSpellingLoc(Tok.getLocation())); 139 } 140 } else if (Tok.getLength() < 256) { 141 char Buffer[256]; 142 const char *TokPtr = Buffer; 143 PP.getSpelling(Tok, TokPtr); 144 return TokPtr[0]; 145 } else { 146 return PP.getSpelling(Tok)[0]; 147 } 148 } 149 150 /// AvoidConcat - If printing PrevTok immediately followed by Tok would cause 151 /// the two individual tokens to be lexed as a single token, return true 152 /// (which causes a space to be printed between them). This allows the output 153 /// of -E mode to be lexed to the same token stream as lexing the input 154 /// directly would. 155 /// 156 /// This code must conservatively return true if it doesn't want to be 100% 157 /// accurate. This will cause the output to include extra space characters, 158 /// but the resulting output won't have incorrect concatenations going on. 159 /// Examples include "..", which we print with a space between, because we 160 /// don't want to track enough to tell "x.." from "...". 161 bool TokenConcatenation::AvoidConcat(const Token &PrevPrevTok, 162 const Token &PrevTok, 163 const Token &Tok) const { 164 // First, check to see if the tokens were directly adjacent in the original 165 // source. If they were, it must be okay to stick them together: if there 166 // were an issue, the tokens would have been lexed differently. 167 SourceManager &SM = PP.getSourceManager(); 168 SourceLocation PrevSpellLoc = SM.getSpellingLoc(PrevTok.getLocation()); 169 SourceLocation SpellLoc = SM.getSpellingLoc(Tok.getLocation()); 170 if (PrevSpellLoc.getLocWithOffset(PrevTok.getLength()) == SpellLoc) 171 return false; 172 173 tok::TokenKind PrevKind = PrevTok.getKind(); 174 if (!PrevTok.isAnnotation() && PrevTok.getIdentifierInfo()) 175 PrevKind = tok::identifier; // Language keyword or named operator. 176 177 // Look up information on when we should avoid concatenation with prevtok. 178 unsigned ConcatInfo = TokenInfo[PrevKind]; 179 180 // If prevtok never causes a problem for anything after it, return quickly. 181 if (ConcatInfo == 0) return false; 182 183 if (ConcatInfo & aci_avoid_equal) { 184 // If the next token is '=' or '==', avoid concatenation. 185 if (Tok.isOneOf(tok::equal, tok::equalequal)) 186 return true; 187 ConcatInfo &= ~aci_avoid_equal; 188 } 189 if (Tok.isAnnotation()) { 190 // Modules annotation can show up when generated automatically for includes. 191 assert(Tok.isOneOf(tok::annot_module_include, tok::annot_module_begin, 192 tok::annot_module_end) && 193 "unexpected annotation in AvoidConcat"); 194 ConcatInfo = 0; 195 } 196 197 if (ConcatInfo == 0) 198 return false; 199 200 // Basic algorithm: we look at the first character of the second token, and 201 // determine whether it, if appended to the first token, would form (or 202 // would contribute) to a larger token if concatenated. 203 char FirstChar = 0; 204 if (ConcatInfo & aci_custom) { 205 // If the token does not need to know the first character, don't get it. 206 } else { 207 FirstChar = GetFirstChar(PP, Tok); 208 } 209 210 switch (PrevKind) { 211 default: 212 llvm_unreachable("InitAvoidConcatTokenInfo built wrong"); 213 214 case tok::raw_identifier: 215 llvm_unreachable("tok::raw_identifier in non-raw lexing mode!"); 216 217 case tok::string_literal: 218 case tok::wide_string_literal: 219 case tok::utf8_string_literal: 220 case tok::utf16_string_literal: 221 case tok::utf32_string_literal: 222 case tok::char_constant: 223 case tok::wide_char_constant: 224 case tok::utf8_char_constant: 225 case tok::utf16_char_constant: 226 case tok::utf32_char_constant: 227 if (!PP.getLangOpts().CPlusPlus11) 228 return false; 229 230 // In C++11, a string or character literal followed by an identifier is a 231 // single token. 232 if (Tok.getIdentifierInfo()) 233 return true; 234 235 // A ud-suffix is an identifier. If the previous token ends with one, treat 236 // it as an identifier. 237 if (!PrevTok.hasUDSuffix()) 238 return false; 239 LLVM_FALLTHROUGH; 240 case tok::identifier: // id+id or id+number or id+L"foo". 241 // id+'.'... will not append. 242 if (Tok.is(tok::numeric_constant)) 243 return GetFirstChar(PP, Tok) != '.'; 244 245 if (Tok.getIdentifierInfo() || 246 Tok.isOneOf(tok::wide_string_literal, tok::utf8_string_literal, 247 tok::utf16_string_literal, tok::utf32_string_literal, 248 tok::wide_char_constant, tok::utf8_char_constant, 249 tok::utf16_char_constant, tok::utf32_char_constant)) 250 return true; 251 252 // If this isn't identifier + string, we're done. 253 if (Tok.isNot(tok::char_constant) && Tok.isNot(tok::string_literal)) 254 return false; 255 256 // Otherwise, this is a narrow character or string. If the *identifier* 257 // is a literal 'L', 'u8', 'u' or 'U', avoid pasting L "foo" -> L"foo". 258 return IsIdentifierStringPrefix(PrevTok); 259 260 case tok::numeric_constant: 261 return isPreprocessingNumberBody(FirstChar) || 262 FirstChar == '+' || FirstChar == '-'; 263 case tok::period: // ..., .*, .1234 264 return (FirstChar == '.' && PrevPrevTok.is(tok::period)) || 265 isDigit(FirstChar) || 266 (PP.getLangOpts().CPlusPlus && FirstChar == '*'); 267 case tok::amp: // && 268 return FirstChar == '&'; 269 case tok::plus: // ++ 270 return FirstChar == '+'; 271 case tok::minus: // --, ->, ->* 272 return FirstChar == '-' || FirstChar == '>'; 273 case tok::slash: //, /*, // 274 return FirstChar == '*' || FirstChar == '/'; 275 case tok::less: // <<, <<=, <:, <% 276 return FirstChar == '<' || FirstChar == ':' || FirstChar == '%'; 277 case tok::greater: // >>, >>= 278 return FirstChar == '>'; 279 case tok::pipe: // || 280 return FirstChar == '|'; 281 case tok::percent: // %>, %: 282 return FirstChar == '>' || FirstChar == ':'; 283 case tok::colon: // ::, :> 284 return FirstChar == '>' || 285 (PP.getLangOpts().CPlusPlus && FirstChar == ':'); 286 case tok::hash: // ##, #@, %:%: 287 return FirstChar == '#' || FirstChar == '@' || FirstChar == '%'; 288 case tok::arrow: // ->* 289 return PP.getLangOpts().CPlusPlus && FirstChar == '*'; 290 case tok::lessequal: // <=> (C++2a) 291 return PP.getLangOpts().CPlusPlus2a && FirstChar == '>'; 292 } 293 } 294