| 1 | //===- TokenLexer.cpp - Lex from a token stream ---------------------------===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | // |
| 9 | // This file implements the TokenLexer interface. |
| 10 | // |
| 11 | //===----------------------------------------------------------------------===// |
| 12 | |
| 13 | #include "clang/Lex/TokenLexer.h" |
| 14 | #include "clang/Basic/Diagnostic.h" |
| 15 | #include "clang/Basic/IdentifierTable.h" |
| 16 | #include "clang/Basic/LangOptions.h" |
| 17 | #include "clang/Basic/SourceLocation.h" |
| 18 | #include "clang/Basic/SourceManager.h" |
| 19 | #include "clang/Basic/TokenKinds.h" |
| 20 | #include "clang/Lex/LexDiagnostic.h" |
| 21 | #include "clang/Lex/Lexer.h" |
| 22 | #include "clang/Lex/MacroArgs.h" |
| 23 | #include "clang/Lex/MacroInfo.h" |
| 24 | #include "clang/Lex/Preprocessor.h" |
| 25 | #include "clang/Lex/Token.h" |
| 26 | #include "clang/Lex/VariadicMacroSupport.h" |
| 27 | #include "llvm/ADT/ArrayRef.h" |
| 28 | #include "llvm/ADT/STLExtras.h" |
| 29 | #include "llvm/ADT/SmallVector.h" |
| 30 | #include "llvm/ADT/iterator_range.h" |
| 31 | #include <cassert> |
| 32 | #include <cstring> |
| 33 | #include <optional> |
| 34 | |
| 35 | using namespace clang; |
| 36 | |
| 37 | /// Create a TokenLexer for the specified macro with the specified actual |
| 38 | /// arguments. Note that this ctor takes ownership of the ActualArgs pointer. |
| 39 | void TokenLexer::Init(Token &Tok, SourceLocation ELEnd, MacroInfo *MI, |
| 40 | MacroArgs *Actuals) { |
| 41 | // If the client is reusing a TokenLexer, make sure to free any memory |
| 42 | // associated with it. |
| 43 | destroy(); |
| 44 | |
| 45 | Macro = MI; |
| 46 | ActualArgs = Actuals; |
| 47 | CurTokenIdx = 0; |
| 48 | |
| 49 | ExpandLocStart = Tok.getLocation(); |
| 50 | ExpandLocEnd = ELEnd; |
| 51 | AtStartOfLine = Tok.isAtStartOfLine(); |
| 52 | HasLeadingSpace = Tok.hasLeadingSpace(); |
| 53 | NextTokGetsSpace = false; |
| 54 | Tokens = &*Macro->tokens_begin(); |
| 55 | OwnsTokens = false; |
| 56 | DisableMacroExpansion = false; |
| 57 | IsReinject = false; |
| 58 | NumTokens = Macro->tokens_end()-Macro->tokens_begin(); |
| 59 | MacroExpansionStart = SourceLocation(); |
| 60 | LexingCXXModuleDirective = false; |
| 61 | |
| 62 | SourceManager &SM = PP.getSourceManager(); |
| 63 | MacroStartSLocOffset = SM.getNextLocalOffset(); |
| 64 | |
| 65 | if (NumTokens > 0) { |
| 66 | assert(Tokens[0].getLocation().isValid()); |
| 67 | assert((Tokens[0].getLocation().isFileID() || Tokens[0].is(tok::comment)) && |
| 68 | "Macro defined in macro?" ); |
| 69 | assert(ExpandLocStart.isValid()); |
| 70 | |
| 71 | // Reserve a source location entry chunk for the length of the macro |
| 72 | // definition. Tokens that get lexed directly from the definition will |
| 73 | // have their locations pointing inside this chunk. This is to avoid |
| 74 | // creating separate source location entries for each token. |
| 75 | MacroDefStart = SM.getExpansionLoc(Loc: Tokens[0].getLocation()); |
| 76 | MacroDefLength = Macro->getDefinitionLength(SM); |
| 77 | MacroExpansionStart = SM.createExpansionLoc(SpellingLoc: MacroDefStart, |
| 78 | ExpansionLocStart: ExpandLocStart, |
| 79 | ExpansionLocEnd: ExpandLocEnd, |
| 80 | Length: MacroDefLength); |
| 81 | } |
| 82 | |
| 83 | // If this is a function-like macro, expand the arguments and change |
| 84 | // Tokens to point to the expanded tokens. |
| 85 | if (Macro->isFunctionLike() && Macro->getNumParams()) |
| 86 | ExpandFunctionArguments(); |
| 87 | |
| 88 | // Mark the macro as currently disabled, so that it is not recursively |
| 89 | // expanded. The macro must be disabled only after argument pre-expansion of |
| 90 | // function-like macro arguments occurs. |
| 91 | Macro->DisableMacro(); |
| 92 | } |
| 93 | |
| 94 | /// Create a TokenLexer for the specified token stream. This does not |
| 95 | /// take ownership of the specified token vector. |
| 96 | void TokenLexer::Init(const Token *TokArray, unsigned NumToks, |
| 97 | bool disableMacroExpansion, bool ownsTokens, |
| 98 | bool isReinject) { |
| 99 | assert(!isReinject || disableMacroExpansion); |
| 100 | // If the client is reusing a TokenLexer, make sure to free any memory |
| 101 | // associated with it. |
| 102 | destroy(); |
| 103 | |
| 104 | Macro = nullptr; |
| 105 | ActualArgs = nullptr; |
| 106 | Tokens = TokArray; |
| 107 | OwnsTokens = ownsTokens; |
| 108 | DisableMacroExpansion = disableMacroExpansion; |
| 109 | IsReinject = isReinject; |
| 110 | NumTokens = NumToks; |
| 111 | CurTokenIdx = 0; |
| 112 | ExpandLocStart = ExpandLocEnd = SourceLocation(); |
| 113 | AtStartOfLine = false; |
| 114 | HasLeadingSpace = false; |
| 115 | NextTokGetsSpace = false; |
| 116 | MacroExpansionStart = SourceLocation(); |
| 117 | LexingCXXModuleDirective = false; |
| 118 | |
| 119 | // Set HasLeadingSpace/AtStartOfLine so that the first token will be |
| 120 | // returned unmodified. |
| 121 | if (NumToks != 0) { |
| 122 | AtStartOfLine = TokArray[0].isAtStartOfLine(); |
| 123 | HasLeadingSpace = TokArray[0].hasLeadingSpace(); |
| 124 | } |
| 125 | } |
| 126 | |
| 127 | void TokenLexer::destroy() { |
| 128 | // If this was a function-like macro that actually uses its arguments, delete |
| 129 | // the expanded tokens. |
| 130 | if (OwnsTokens) { |
| 131 | delete [] Tokens; |
| 132 | Tokens = nullptr; |
| 133 | OwnsTokens = false; |
| 134 | } |
| 135 | |
| 136 | // TokenLexer owns its formal arguments. |
| 137 | if (ActualArgs) ActualArgs->destroy(PP); |
| 138 | } |
| 139 | |
| 140 | bool TokenLexer::MaybeRemoveCommaBeforeVaArgs( |
| 141 | SmallVectorImpl<Token> &ResultToks, bool HasPasteOperator, MacroInfo *Macro, |
| 142 | unsigned MacroArgNo, Preprocessor &PP) { |
| 143 | // Is the macro argument __VA_ARGS__? |
| 144 | if (!Macro->isVariadic() || MacroArgNo != Macro->getNumParams()-1) |
| 145 | return false; |
| 146 | |
| 147 | // In Microsoft-compatibility mode, a comma is removed in the expansion |
| 148 | // of " ... , __VA_ARGS__ " if __VA_ARGS__ is empty. This extension is |
| 149 | // not supported by gcc. |
| 150 | if (!HasPasteOperator && !PP.getLangOpts().MSVCCompat) |
| 151 | return false; |
| 152 | |
| 153 | // GCC removes the comma in the expansion of " ... , ## __VA_ARGS__ " if |
| 154 | // __VA_ARGS__ is empty, but not in strict C99 mode where there are no |
| 155 | // named arguments, where it remains. In all other modes, including C99 |
| 156 | // with GNU extensions, it is removed regardless of named arguments. |
| 157 | // Microsoft also appears to support this extension, unofficially. |
| 158 | if (PP.getLangOpts().C99 && !PP.getLangOpts().GNUMode |
| 159 | && Macro->getNumParams() < 2) |
| 160 | return false; |
| 161 | |
| 162 | // Is a comma available to be removed? |
| 163 | if (ResultToks.empty() || !ResultToks.back().is(K: tok::comma)) |
| 164 | return false; |
| 165 | |
| 166 | // Issue an extension diagnostic for the paste operator. |
| 167 | if (HasPasteOperator) |
| 168 | PP.Diag(Loc: ResultToks.back().getLocation(), DiagID: diag::ext_paste_comma); |
| 169 | |
| 170 | // Remove the comma. |
| 171 | ResultToks.pop_back(); |
| 172 | |
| 173 | if (!ResultToks.empty()) { |
| 174 | // If the comma was right after another paste (e.g. "X##,##__VA_ARGS__"), |
| 175 | // then removal of the comma should produce a placemarker token (in C99 |
| 176 | // terms) which we model by popping off the previous ##, giving us a plain |
| 177 | // "X" when __VA_ARGS__ is empty. |
| 178 | if (ResultToks.back().is(K: tok::hashhash)) |
| 179 | ResultToks.pop_back(); |
| 180 | |
| 181 | // Remember that this comma was elided. |
| 182 | ResultToks.back().setFlag(Token::CommaAfterElided); |
| 183 | } |
| 184 | |
| 185 | // Never add a space, even if the comma, ##, or arg had a space. |
| 186 | NextTokGetsSpace = false; |
| 187 | return true; |
| 188 | } |
| 189 | |
| 190 | void TokenLexer::stringifyVAOPTContents( |
| 191 | SmallVectorImpl<Token> &ResultToks, const VAOptExpansionContext &VCtx, |
| 192 | const SourceLocation VAOPTClosingParenLoc) { |
| 193 | const int NumToksPriorToVAOpt = VCtx.getNumberOfTokensPriorToVAOpt(); |
| 194 | const unsigned int NumVAOptTokens = ResultToks.size() - NumToksPriorToVAOpt; |
| 195 | Token *const VAOPTTokens = |
| 196 | NumVAOptTokens ? &ResultToks[NumToksPriorToVAOpt] : nullptr; |
| 197 | |
| 198 | SmallVector<Token, 64> ConcatenatedVAOPTResultToks; |
| 199 | // FIXME: Should we keep track within VCtx that we did or didnot |
| 200 | // encounter pasting - and only then perform this loop. |
| 201 | |
| 202 | // Perform token pasting (concatenation) prior to stringization. |
| 203 | for (unsigned int CurTokenIdx = 0; CurTokenIdx != NumVAOptTokens; |
| 204 | ++CurTokenIdx) { |
| 205 | if (VAOPTTokens[CurTokenIdx].is(K: tok::hashhash)) { |
| 206 | assert(CurTokenIdx != 0 && |
| 207 | "Can not have __VAOPT__ contents begin with a ##" ); |
| 208 | Token &LHS = VAOPTTokens[CurTokenIdx - 1]; |
| 209 | pasteTokens(LHSTok&: LHS, TokenStream: llvm::ArrayRef(VAOPTTokens, NumVAOptTokens), |
| 210 | CurIdx&: CurTokenIdx); |
| 211 | // Replace the token prior to the first ## in this iteration. |
| 212 | ConcatenatedVAOPTResultToks.back() = LHS; |
| 213 | if (CurTokenIdx == NumVAOptTokens) |
| 214 | break; |
| 215 | } |
| 216 | ConcatenatedVAOPTResultToks.push_back(Elt: VAOPTTokens[CurTokenIdx]); |
| 217 | } |
| 218 | |
| 219 | ConcatenatedVAOPTResultToks.push_back(Elt: VCtx.getEOFTok()); |
| 220 | // Get the SourceLocation that represents the start location within |
| 221 | // the macro definition that marks where this string is substituted |
| 222 | // into: i.e. the __VA_OPT__ and the ')' within the spelling of the |
| 223 | // macro definition, and use it to indicate that the stringified token |
| 224 | // was generated from that location. |
| 225 | const SourceLocation ExpansionLocStartWithinMacro = |
| 226 | getExpansionLocForMacroDefLoc(loc: VCtx.getVAOptLoc()); |
| 227 | const SourceLocation ExpansionLocEndWithinMacro = |
| 228 | getExpansionLocForMacroDefLoc(loc: VAOPTClosingParenLoc); |
| 229 | |
| 230 | Token StringifiedVAOPT = MacroArgs::StringifyArgument( |
| 231 | ArgToks: &ConcatenatedVAOPTResultToks[0], PP, Charify: VCtx.hasCharifyBefore() /*Charify*/, |
| 232 | ExpansionLocStart: ExpansionLocStartWithinMacro, ExpansionLocEnd: ExpansionLocEndWithinMacro); |
| 233 | |
| 234 | if (VCtx.getLeadingSpaceForStringifiedToken()) |
| 235 | StringifiedVAOPT.setFlag(Token::LeadingSpace); |
| 236 | |
| 237 | StringifiedVAOPT.setFlag(Token::StringifiedInMacro); |
| 238 | // Resize (shrink) the token stream to just capture this stringified token. |
| 239 | ResultToks.resize(N: NumToksPriorToVAOpt + 1); |
| 240 | ResultToks.back() = StringifiedVAOPT; |
| 241 | } |
| 242 | |
| 243 | /// Expand the arguments of a function-like macro so that we can quickly |
| 244 | /// return preexpanded tokens from Tokens. |
| 245 | void TokenLexer::ExpandFunctionArguments() { |
| 246 | SmallVector<Token, 128> ResultToks; |
| 247 | |
| 248 | // Loop through 'Tokens', expanding them into ResultToks. Keep |
| 249 | // track of whether we change anything. If not, no need to keep them. If so, |
| 250 | // we install the newly expanded sequence as the new 'Tokens' list. |
| 251 | bool MadeChange = false; |
| 252 | |
| 253 | std::optional<bool> CalledWithVariadicArguments; |
| 254 | |
| 255 | VAOptExpansionContext VCtx(PP); |
| 256 | |
| 257 | for (unsigned I = 0, E = NumTokens; I != E; ++I) { |
| 258 | const Token &CurTok = Tokens[I]; |
| 259 | // We don't want a space for the next token after a paste |
| 260 | // operator. In valid code, the token will get smooshed onto the |
| 261 | // preceding one anyway. In assembler-with-cpp mode, invalid |
| 262 | // pastes are allowed through: in this case, we do not want the |
| 263 | // extra whitespace to be added. For example, we want ". ## foo" |
| 264 | // -> ".foo" not ". foo". |
| 265 | if (I != 0 && !Tokens[I-1].is(K: tok::hashhash) && CurTok.hasLeadingSpace()) |
| 266 | NextTokGetsSpace = true; |
| 267 | |
| 268 | if (VCtx.isVAOptToken(T: CurTok)) { |
| 269 | MadeChange = true; |
| 270 | assert(Tokens[I + 1].is(tok::l_paren) && |
| 271 | "__VA_OPT__ must be followed by '('" ); |
| 272 | |
| 273 | ++I; // Skip the l_paren |
| 274 | VCtx.sawVAOptFollowedByOpeningParens(VAOptLoc: CurTok.getLocation(), |
| 275 | NumPriorTokens: ResultToks.size()); |
| 276 | |
| 277 | continue; |
| 278 | } |
| 279 | |
| 280 | // We have entered into the __VA_OPT__ context, so handle tokens |
| 281 | // appropriately. |
| 282 | if (VCtx.isInVAOpt()) { |
| 283 | // If we are about to process a token that is either an argument to |
| 284 | // __VA_OPT__ or its closing rparen, then: |
| 285 | // 1) If the token is the closing rparen that exits us out of __VA_OPT__, |
| 286 | // perform any necessary stringification or placemarker processing, |
| 287 | // and/or skip to the next token. |
| 288 | // 2) else if macro was invoked without variadic arguments skip this |
| 289 | // token. |
| 290 | // 3) else (macro was invoked with variadic arguments) process the token |
| 291 | // normally. |
| 292 | |
| 293 | if (Tokens[I].is(K: tok::l_paren)) |
| 294 | VCtx.sawOpeningParen(LParenLoc: Tokens[I].getLocation()); |
| 295 | // Continue skipping tokens within __VA_OPT__ if the macro was not |
| 296 | // called with variadic arguments, else let the rest of the loop handle |
| 297 | // this token. Note sawClosingParen() returns true only if the r_paren matches |
| 298 | // the closing r_paren of the __VA_OPT__. |
| 299 | if (!Tokens[I].is(K: tok::r_paren) || !VCtx.sawClosingParen()) { |
| 300 | // Lazily expand __VA_ARGS__ when we see the first __VA_OPT__. |
| 301 | if (!CalledWithVariadicArguments) { |
| 302 | CalledWithVariadicArguments = |
| 303 | ActualArgs->invokedWithVariadicArgument(MI: Macro, PP); |
| 304 | } |
| 305 | if (!*CalledWithVariadicArguments) { |
| 306 | // Skip this token. |
| 307 | continue; |
| 308 | } |
| 309 | // ... else the macro was called with variadic arguments, and we do not |
| 310 | // have a closing rparen - so process this token normally. |
| 311 | } else { |
| 312 | // Current token is the closing r_paren which marks the end of the |
| 313 | // __VA_OPT__ invocation, so handle any place-marker pasting (if |
| 314 | // empty) by removing hashhash either before (if exists) or after. And |
| 315 | // also stringify the entire contents if VAOPT was preceded by a hash, |
| 316 | // but do so only after any token concatenation that needs to occur |
| 317 | // within the contents of VAOPT. |
| 318 | |
| 319 | if (VCtx.hasStringifyOrCharifyBefore()) { |
| 320 | // Replace all the tokens just added from within VAOPT into a single |
| 321 | // stringified token. This requires token-pasting to eagerly occur |
| 322 | // within these tokens. If either the contents of VAOPT were empty |
| 323 | // or the macro wasn't called with any variadic arguments, the result |
| 324 | // is a token that represents an empty string. |
| 325 | stringifyVAOPTContents(ResultToks, VCtx, |
| 326 | /*ClosingParenLoc*/ VAOPTClosingParenLoc: Tokens[I].getLocation()); |
| 327 | |
| 328 | } else if (/*No tokens within VAOPT*/ |
| 329 | ResultToks.size() == VCtx.getNumberOfTokensPriorToVAOpt()) { |
| 330 | // Treat VAOPT as a placemarker token. Eat either the '##' before the |
| 331 | // RHS/VAOPT (if one exists, suggesting that the LHS (if any) to that |
| 332 | // hashhash was not a placemarker) or the '##' |
| 333 | // after VAOPT, but not both. |
| 334 | |
| 335 | if (ResultToks.size() && ResultToks.back().is(K: tok::hashhash)) { |
| 336 | ResultToks.pop_back(); |
| 337 | } else if ((I + 1 != E) && Tokens[I + 1].is(K: tok::hashhash)) { |
| 338 | ++I; // Skip the following hashhash. |
| 339 | } |
| 340 | } else { |
| 341 | // If there's a ## before the __VA_OPT__, we might have discovered |
| 342 | // that the __VA_OPT__ begins with a placeholder. We delay action on |
| 343 | // that to now to avoid messing up our stashed count of tokens before |
| 344 | // __VA_OPT__. |
| 345 | if (VCtx.beginsWithPlaceholder()) { |
| 346 | assert(VCtx.getNumberOfTokensPriorToVAOpt() > 0 && |
| 347 | ResultToks.size() >= VCtx.getNumberOfTokensPriorToVAOpt() && |
| 348 | ResultToks[VCtx.getNumberOfTokensPriorToVAOpt() - 1].is( |
| 349 | tok::hashhash) && |
| 350 | "no token paste before __VA_OPT__" ); |
| 351 | ResultToks.erase(CI: ResultToks.begin() + |
| 352 | VCtx.getNumberOfTokensPriorToVAOpt() - 1); |
| 353 | } |
| 354 | // If the expansion of __VA_OPT__ ends with a placeholder, eat any |
| 355 | // following '##' token. |
| 356 | if (VCtx.endsWithPlaceholder() && I + 1 != E && |
| 357 | Tokens[I + 1].is(K: tok::hashhash)) { |
| 358 | ++I; |
| 359 | } |
| 360 | } |
| 361 | VCtx.reset(); |
| 362 | // We processed __VA_OPT__'s closing paren (and the exit out of |
| 363 | // __VA_OPT__), so skip to the next token. |
| 364 | continue; |
| 365 | } |
| 366 | } |
| 367 | |
| 368 | // If we found the stringify operator, get the argument stringified. The |
| 369 | // preprocessor already verified that the following token is a macro |
| 370 | // parameter or __VA_OPT__ when the #define was lexed. |
| 371 | |
| 372 | if (CurTok.isOneOf(Ks: tok::hash, Ks: tok::hashat)) { |
| 373 | int ArgNo = Macro->getParameterNum(Arg: Tokens[I+1].getIdentifierInfo()); |
| 374 | assert((ArgNo != -1 || VCtx.isVAOptToken(Tokens[I + 1])) && |
| 375 | "Token following # is not an argument or __VA_OPT__!" ); |
| 376 | |
| 377 | if (ArgNo == -1) { |
| 378 | // Handle the __VA_OPT__ case. |
| 379 | VCtx.sawHashOrHashAtBefore(HasLeadingSpace: NextTokGetsSpace, |
| 380 | IsHashAt: CurTok.is(K: tok::hashat)); |
| 381 | continue; |
| 382 | } |
| 383 | // Else handle the simple argument case. |
| 384 | SourceLocation ExpansionLocStart = |
| 385 | getExpansionLocForMacroDefLoc(loc: CurTok.getLocation()); |
| 386 | SourceLocation ExpansionLocEnd = |
| 387 | getExpansionLocForMacroDefLoc(loc: Tokens[I+1].getLocation()); |
| 388 | |
| 389 | bool Charify = CurTok.is(K: tok::hashat); |
| 390 | const Token *UnexpArg = ActualArgs->getUnexpArgument(Arg: ArgNo); |
| 391 | Token Res = MacroArgs::StringifyArgument( |
| 392 | ArgToks: UnexpArg, PP, Charify, ExpansionLocStart, ExpansionLocEnd); |
| 393 | Res.setFlag(Token::StringifiedInMacro); |
| 394 | |
| 395 | // The stringified/charified string leading space flag gets set to match |
| 396 | // the #/#@ operator. |
| 397 | if (NextTokGetsSpace) |
| 398 | Res.setFlag(Token::LeadingSpace); |
| 399 | |
| 400 | ResultToks.push_back(Elt: Res); |
| 401 | MadeChange = true; |
| 402 | ++I; // Skip arg name. |
| 403 | NextTokGetsSpace = false; |
| 404 | continue; |
| 405 | } |
| 406 | |
| 407 | // Find out if there is a paste (##) operator before or after the token. |
| 408 | bool NonEmptyPasteBefore = |
| 409 | !ResultToks.empty() && ResultToks.back().is(K: tok::hashhash); |
| 410 | bool PasteBefore = I != 0 && Tokens[I-1].is(K: tok::hashhash); |
| 411 | bool PasteAfter = I+1 != E && Tokens[I+1].is(K: tok::hashhash); |
| 412 | bool RParenAfter = I+1 != E && Tokens[I+1].is(K: tok::r_paren); |
| 413 | |
| 414 | assert((!NonEmptyPasteBefore || PasteBefore || VCtx.isInVAOpt()) && |
| 415 | "unexpected ## in ResultToks" ); |
| 416 | |
| 417 | // Otherwise, if this is not an argument token, just add the token to the |
| 418 | // output buffer. |
| 419 | IdentifierInfo *II = CurTok.getIdentifierInfo(); |
| 420 | int ArgNo = II ? Macro->getParameterNum(Arg: II) : -1; |
| 421 | if (ArgNo == -1) { |
| 422 | // This isn't an argument, just add it. |
| 423 | ResultToks.push_back(Elt: CurTok); |
| 424 | |
| 425 | if (NextTokGetsSpace) { |
| 426 | ResultToks.back().setFlag(Token::LeadingSpace); |
| 427 | NextTokGetsSpace = false; |
| 428 | } else if (PasteBefore && !NonEmptyPasteBefore) |
| 429 | ResultToks.back().clearFlag(Flag: Token::LeadingSpace); |
| 430 | |
| 431 | continue; |
| 432 | } |
| 433 | |
| 434 | // An argument is expanded somehow, the result is different than the |
| 435 | // input. |
| 436 | MadeChange = true; |
| 437 | |
| 438 | // Otherwise, this is a use of the argument. |
| 439 | |
| 440 | // In Microsoft mode, remove the comma before __VA_ARGS__ to ensure there |
| 441 | // are no trailing commas if __VA_ARGS__ is empty. |
| 442 | if (!PasteBefore && ActualArgs->isVarargsElidedUse() && |
| 443 | MaybeRemoveCommaBeforeVaArgs(ResultToks, |
| 444 | /*HasPasteOperator=*/false, |
| 445 | Macro, MacroArgNo: ArgNo, PP)) |
| 446 | continue; |
| 447 | |
| 448 | // If it is not the LHS/RHS of a ## operator, we must pre-expand the |
| 449 | // argument and substitute the expanded tokens into the result. This is |
| 450 | // C99 6.10.3.1p1. |
| 451 | if (!PasteBefore && !PasteAfter) { |
| 452 | const Token *ResultArgToks; |
| 453 | |
| 454 | // Only preexpand the argument if it could possibly need it. This |
| 455 | // avoids some work in common cases. |
| 456 | const Token *ArgTok = ActualArgs->getUnexpArgument(Arg: ArgNo); |
| 457 | if (ActualArgs->ArgNeedsPreexpansion(ArgTok, PP)) |
| 458 | ResultArgToks = &ActualArgs->getPreExpArgument(Arg: ArgNo, PP)[0]; |
| 459 | else |
| 460 | ResultArgToks = ArgTok; // Use non-preexpanded tokens. |
| 461 | |
| 462 | // If the arg token expanded into anything, append it. |
| 463 | if (ResultArgToks->isNot(K: tok::eof)) { |
| 464 | size_t FirstResult = ResultToks.size(); |
| 465 | unsigned NumToks = MacroArgs::getArgLength(ArgPtr: ResultArgToks); |
| 466 | ResultToks.append(in_start: ResultArgToks, in_end: ResultArgToks+NumToks); |
| 467 | |
| 468 | // In Microsoft-compatibility mode, we follow MSVC's preprocessing |
| 469 | // behavior by not considering single commas from nested macro |
| 470 | // expansions as argument separators. Set a flag on the token so we can |
| 471 | // test for this later when the macro expansion is processed. |
| 472 | if (PP.getLangOpts().MSVCCompat && NumToks == 1 && |
| 473 | ResultToks.back().is(K: tok::comma)) |
| 474 | ResultToks.back().setFlag(Token::IgnoredComma); |
| 475 | |
| 476 | // If the '##' came from expanding an argument, turn it into 'unknown' |
| 477 | // to avoid pasting. |
| 478 | for (Token &Tok : llvm::drop_begin(RangeOrContainer&: ResultToks, N: FirstResult)) |
| 479 | if (Tok.is(K: tok::hashhash)) |
| 480 | Tok.setKind(tok::unknown); |
| 481 | |
| 482 | if(ExpandLocStart.isValid()) { |
| 483 | updateLocForMacroArgTokens(ArgIdSpellLoc: CurTok.getLocation(), |
| 484 | begin_tokens: ResultToks.begin()+FirstResult, |
| 485 | end_tokens: ResultToks.end()); |
| 486 | } |
| 487 | |
| 488 | // If any tokens were substituted from the argument, the whitespace |
| 489 | // before the first token should match the whitespace of the arg |
| 490 | // identifier. |
| 491 | ResultToks[FirstResult].setFlagValue(Flag: Token::LeadingSpace, |
| 492 | Val: NextTokGetsSpace); |
| 493 | ResultToks[FirstResult].setFlagValue(Flag: Token::StartOfLine, Val: false); |
| 494 | NextTokGetsSpace = false; |
| 495 | } else { |
| 496 | // We're creating a placeholder token. Usually this doesn't matter, |
| 497 | // but it can affect paste behavior when at the start or end of a |
| 498 | // __VA_OPT__. |
| 499 | if (NonEmptyPasteBefore) { |
| 500 | // We're imagining a placeholder token is inserted here. If this is |
| 501 | // the first token in a __VA_OPT__ after a ##, delete the ##. |
| 502 | assert(VCtx.isInVAOpt() && "should only happen inside a __VA_OPT__" ); |
| 503 | VCtx.hasPlaceholderAfterHashhashAtStart(); |
| 504 | } else if (RParenAfter) |
| 505 | VCtx.hasPlaceholderBeforeRParen(); |
| 506 | } |
| 507 | continue; |
| 508 | } |
| 509 | |
| 510 | // Okay, we have a token that is either the LHS or RHS of a paste (##) |
| 511 | // argument. It gets substituted as its non-pre-expanded tokens. |
| 512 | const Token *ArgToks = ActualArgs->getUnexpArgument(Arg: ArgNo); |
| 513 | unsigned NumToks = MacroArgs::getArgLength(ArgPtr: ArgToks); |
| 514 | if (NumToks) { // Not an empty argument? |
| 515 | bool VaArgsPseudoPaste = false; |
| 516 | // If this is the GNU ", ## __VA_ARGS__" extension, and we just learned |
| 517 | // that __VA_ARGS__ expands to multiple tokens, avoid a pasting error when |
| 518 | // the expander tries to paste ',' with the first token of the __VA_ARGS__ |
| 519 | // expansion. |
| 520 | if (NonEmptyPasteBefore && ResultToks.size() >= 2 && |
| 521 | ResultToks[ResultToks.size()-2].is(K: tok::comma) && |
| 522 | (unsigned)ArgNo == Macro->getNumParams()-1 && |
| 523 | Macro->isVariadic()) { |
| 524 | VaArgsPseudoPaste = true; |
| 525 | // Remove the paste operator, report use of the extension. |
| 526 | PP.Diag(Loc: ResultToks.pop_back_val().getLocation(), DiagID: diag::ext_paste_comma); |
| 527 | } |
| 528 | |
| 529 | ResultToks.append(in_start: ArgToks, in_end: ArgToks+NumToks); |
| 530 | |
| 531 | // If the '##' came from expanding an argument, turn it into 'unknown' |
| 532 | // to avoid pasting. |
| 533 | for (Token &Tok : llvm::make_range(x: ResultToks.end() - NumToks, |
| 534 | y: ResultToks.end())) { |
| 535 | if (Tok.is(K: tok::hashhash)) |
| 536 | Tok.setKind(tok::unknown); |
| 537 | } |
| 538 | |
| 539 | if (ExpandLocStart.isValid()) { |
| 540 | updateLocForMacroArgTokens(ArgIdSpellLoc: CurTok.getLocation(), |
| 541 | begin_tokens: ResultToks.end()-NumToks, end_tokens: ResultToks.end()); |
| 542 | } |
| 543 | |
| 544 | // Transfer the leading whitespace information from the token |
| 545 | // (the macro argument) onto the first token of the |
| 546 | // expansion. Note that we don't do this for the GNU |
| 547 | // pseudo-paste extension ", ## __VA_ARGS__". |
| 548 | if (!VaArgsPseudoPaste) { |
| 549 | ResultToks[ResultToks.size() - NumToks].setFlagValue(Flag: Token::StartOfLine, |
| 550 | Val: false); |
| 551 | ResultToks[ResultToks.size() - NumToks].setFlagValue( |
| 552 | Flag: Token::LeadingSpace, Val: NextTokGetsSpace); |
| 553 | } |
| 554 | |
| 555 | NextTokGetsSpace = false; |
| 556 | continue; |
| 557 | } |
| 558 | |
| 559 | // If an empty argument is on the LHS or RHS of a paste, the standard (C99 |
| 560 | // 6.10.3.3p2,3) calls for a bunch of placemarker stuff to occur. We |
| 561 | // implement this by eating ## operators when a LHS or RHS expands to |
| 562 | // empty. |
| 563 | if (PasteAfter) { |
| 564 | // Discard the argument token and skip (don't copy to the expansion |
| 565 | // buffer) the paste operator after it. |
| 566 | ++I; |
| 567 | continue; |
| 568 | } |
| 569 | |
| 570 | if (RParenAfter && !NonEmptyPasteBefore) |
| 571 | VCtx.hasPlaceholderBeforeRParen(); |
| 572 | |
| 573 | // If this is on the RHS of a paste operator, we've already copied the |
| 574 | // paste operator to the ResultToks list, unless the LHS was empty too. |
| 575 | // Remove it. |
| 576 | assert(PasteBefore); |
| 577 | if (NonEmptyPasteBefore) { |
| 578 | assert(ResultToks.back().is(tok::hashhash)); |
| 579 | // Do not remove the paste operator if it is the one before __VA_OPT__ |
| 580 | // (and we are still processing tokens within VA_OPT). We handle the case |
| 581 | // of removing the paste operator if __VA_OPT__ reduces to the notional |
| 582 | // placemarker above when we encounter the closing paren of VA_OPT. |
| 583 | if (!VCtx.isInVAOpt() || |
| 584 | ResultToks.size() > VCtx.getNumberOfTokensPriorToVAOpt()) |
| 585 | ResultToks.pop_back(); |
| 586 | else |
| 587 | VCtx.hasPlaceholderAfterHashhashAtStart(); |
| 588 | } |
| 589 | |
| 590 | // If this is the __VA_ARGS__ token, and if the argument wasn't provided, |
| 591 | // and if the macro had at least one real argument, and if the token before |
| 592 | // the ## was a comma, remove the comma. This is a GCC extension which is |
| 593 | // disabled when using -std=c99. |
| 594 | if (ActualArgs->isVarargsElidedUse()) |
| 595 | MaybeRemoveCommaBeforeVaArgs(ResultToks, |
| 596 | /*HasPasteOperator=*/true, |
| 597 | Macro, MacroArgNo: ArgNo, PP); |
| 598 | } |
| 599 | |
| 600 | // If anything changed, install this as the new Tokens list. |
| 601 | if (MadeChange) { |
| 602 | assert(!OwnsTokens && "This would leak if we already own the token list" ); |
| 603 | // This is deleted in the dtor. |
| 604 | NumTokens = ResultToks.size(); |
| 605 | // The tokens will be added to Preprocessor's cache and will be removed |
| 606 | // when this TokenLexer finishes lexing them. |
| 607 | Tokens = PP.cacheMacroExpandedTokens(tokLexer: this, tokens: ResultToks); |
| 608 | |
| 609 | // The preprocessor cache of macro expanded tokens owns these tokens,not us. |
| 610 | OwnsTokens = false; |
| 611 | } |
| 612 | } |
| 613 | |
| 614 | /// Checks if two tokens form wide string literal. |
| 615 | static bool isWideStringLiteralFromMacro(const Token &FirstTok, |
| 616 | const Token &SecondTok) { |
| 617 | return FirstTok.is(K: tok::identifier) && |
| 618 | FirstTok.getIdentifierInfo()->isStr(Str: "L" ) && SecondTok.isLiteral() && |
| 619 | SecondTok.stringifiedInMacro(); |
| 620 | } |
| 621 | |
| 622 | /// Lex - Lex and return a token from this macro stream. |
| 623 | bool TokenLexer::Lex(Token &Tok) { |
| 624 | // Lexing off the end of the macro, pop this macro off the expansion stack. |
| 625 | if (isAtEnd()) { |
| 626 | // If this is a macro (not a token stream), mark the macro enabled now |
| 627 | // that it is no longer being expanded. |
| 628 | if (Macro) Macro->EnableMacro(); |
| 629 | |
| 630 | // CWG2947: Allow the following code: |
| 631 | // |
| 632 | // export module m; int x; |
| 633 | // extern "C++" int *y = &x; |
| 634 | // |
| 635 | // The 'extern' token should has 'StartOfLine' flag when current TokenLexer |
| 636 | // exits and propagate line start/leading space info. |
| 637 | if (!Macro && isLexingCXXModuleDirective()) { |
| 638 | AtStartOfLine = true; |
| 639 | setLexingCXXModuleDirective(false); |
| 640 | } |
| 641 | |
| 642 | Tok.startToken(); |
| 643 | Tok.setFlagValue(Flag: Token::StartOfLine , Val: AtStartOfLine); |
| 644 | Tok.setFlagValue(Flag: Token::LeadingSpace, Val: HasLeadingSpace || NextTokGetsSpace); |
| 645 | if (CurTokenIdx == 0) |
| 646 | Tok.setFlag(Token::LeadingEmptyMacro); |
| 647 | return PP.HandleEndOfTokenLexer(Result&: Tok); |
| 648 | } |
| 649 | |
| 650 | SourceManager &SM = PP.getSourceManager(); |
| 651 | |
| 652 | // If this is the first token of the expanded result, we inherit spacing |
| 653 | // properties later. |
| 654 | bool isFirstToken = CurTokenIdx == 0; |
| 655 | |
| 656 | // Get the next token to return. |
| 657 | Tok = Tokens[CurTokenIdx++]; |
| 658 | if (IsReinject) |
| 659 | Tok.setFlag(Token::IsReinjected); |
| 660 | |
| 661 | bool TokenIsFromPaste = false; |
| 662 | |
| 663 | // If this token is followed by a token paste (##) operator, paste the tokens! |
| 664 | // Note that ## is a normal token when not expanding a macro. |
| 665 | if (!isAtEnd() && Macro && |
| 666 | (Tokens[CurTokenIdx].is(K: tok::hashhash) || |
| 667 | // Special processing of L#x macros in -fms-compatibility mode. |
| 668 | // Microsoft compiler is able to form a wide string literal from |
| 669 | // 'L#macro_arg' construct in a function-like macro. |
| 670 | (PP.getLangOpts().MSVCCompat && |
| 671 | isWideStringLiteralFromMacro(FirstTok: Tok, SecondTok: Tokens[CurTokenIdx])))) { |
| 672 | // When handling the microsoft /##/ extension, the final token is |
| 673 | // returned by pasteTokens, not the pasted token. |
| 674 | if (pasteTokens(Tok)) |
| 675 | return true; |
| 676 | |
| 677 | TokenIsFromPaste = true; |
| 678 | } |
| 679 | |
| 680 | // The token's current location indicate where the token was lexed from. We |
| 681 | // need this information to compute the spelling of the token, but any |
| 682 | // diagnostics for the expanded token should appear as if they came from |
| 683 | // ExpansionLoc. Pull this information together into a new SourceLocation |
| 684 | // that captures all of this. |
| 685 | if (ExpandLocStart.isValid() && // Don't do this for token streams. |
| 686 | // Check that the token's location was not already set properly. |
| 687 | SM.isBeforeInSLocAddrSpace(LHS: Tok.getLocation(), RHS: MacroStartSLocOffset)) { |
| 688 | SourceLocation instLoc; |
| 689 | if (Tok.is(K: tok::comment)) { |
| 690 | instLoc = SM.createExpansionLoc(SpellingLoc: Tok.getLocation(), |
| 691 | ExpansionLocStart: ExpandLocStart, |
| 692 | ExpansionLocEnd: ExpandLocEnd, |
| 693 | Length: Tok.getLength()); |
| 694 | } else { |
| 695 | instLoc = getExpansionLocForMacroDefLoc(loc: Tok.getLocation()); |
| 696 | } |
| 697 | |
| 698 | Tok.setLocation(instLoc); |
| 699 | } |
| 700 | |
| 701 | // If this is the first token, set the lexical properties of the token to |
| 702 | // match the lexical properties of the macro identifier. |
| 703 | if (isFirstToken) { |
| 704 | Tok.setFlagValue(Flag: Token::StartOfLine , Val: AtStartOfLine); |
| 705 | Tok.setFlagValue(Flag: Token::LeadingSpace, Val: HasLeadingSpace); |
| 706 | } else { |
| 707 | // If this is not the first token, we may still need to pass through |
| 708 | // leading whitespace if we've expanded a macro. |
| 709 | if (AtStartOfLine) Tok.setFlag(Token::StartOfLine); |
| 710 | if (HasLeadingSpace) Tok.setFlag(Token::LeadingSpace); |
| 711 | } |
| 712 | AtStartOfLine = false; |
| 713 | HasLeadingSpace = false; |
| 714 | |
| 715 | // Handle recursive expansion! |
| 716 | if (!Tok.isAnnotation() && Tok.getIdentifierInfo() != nullptr && |
| 717 | (!PP.getLangOpts().CPlusPlusModules || |
| 718 | !Tok.isModuleContextualKeyword())) { |
| 719 | // Change the kind of this identifier to the appropriate token kind, e.g. |
| 720 | // turning "for" into a keyword. |
| 721 | IdentifierInfo *II = Tok.getIdentifierInfo(); |
| 722 | Tok.setKind(II->getTokenID()); |
| 723 | |
| 724 | // If this identifier was poisoned and from a paste, emit an error. This |
| 725 | // won't be handled by Preprocessor::HandleIdentifier because this is coming |
| 726 | // from a macro expansion. |
| 727 | if (II->isPoisoned() && TokenIsFromPaste) { |
| 728 | PP.HandlePoisonedIdentifier(Identifier&: Tok); |
| 729 | } |
| 730 | |
| 731 | if (!DisableMacroExpansion && II->isHandleIdentifierCase()) |
| 732 | return PP.HandleIdentifier(Identifier&: Tok); |
| 733 | } |
| 734 | |
| 735 | // Otherwise, return a normal token. |
| 736 | return true; |
| 737 | } |
| 738 | |
| 739 | bool TokenLexer::pasteTokens(Token &Tok) { |
| 740 | return pasteTokens(LHSTok&: Tok, TokenStream: llvm::ArrayRef(Tokens, NumTokens), CurIdx&: CurTokenIdx); |
| 741 | } |
| 742 | |
| 743 | /// LHSTok is the LHS of a ## operator, and CurTokenIdx is the ## |
| 744 | /// operator. Read the ## and RHS, and paste the LHS/RHS together. If there |
| 745 | /// are more ## after it, chomp them iteratively. Return the result as LHSTok. |
| 746 | /// If this returns true, the caller should immediately return the token. |
| 747 | bool TokenLexer::pasteTokens(Token &LHSTok, ArrayRef<Token> TokenStream, |
| 748 | unsigned int &CurIdx) { |
| 749 | assert(CurIdx > 0 && "## can not be the first token within tokens" ); |
| 750 | assert((TokenStream[CurIdx].is(tok::hashhash) || |
| 751 | (PP.getLangOpts().MSVCCompat && |
| 752 | isWideStringLiteralFromMacro(LHSTok, TokenStream[CurIdx]))) && |
| 753 | "Token at this Index must be ## or part of the MSVC 'L " |
| 754 | "#macro-arg' pasting pair" ); |
| 755 | |
| 756 | // MSVC: If previous token was pasted, this must be a recovery from an invalid |
| 757 | // paste operation. Ignore spaces before this token to mimic MSVC output. |
| 758 | // Required for generating valid UUID strings in some MS headers. |
| 759 | if (PP.getLangOpts().MicrosoftExt && (CurIdx >= 2) && |
| 760 | TokenStream[CurIdx - 2].is(K: tok::hashhash)) |
| 761 | LHSTok.clearFlag(Flag: Token::LeadingSpace); |
| 762 | |
| 763 | SmallString<128> Buffer; |
| 764 | const char *ResultTokStrPtr = nullptr; |
| 765 | SourceLocation StartLoc = LHSTok.getLocation(); |
| 766 | SourceLocation PasteOpLoc; |
| 767 | bool HasUCNs = false; |
| 768 | |
| 769 | auto IsAtEnd = [&TokenStream, &CurIdx] { |
| 770 | return TokenStream.size() == CurIdx; |
| 771 | }; |
| 772 | |
| 773 | do { |
| 774 | // Consume the ## operator if any. |
| 775 | PasteOpLoc = TokenStream[CurIdx].getLocation(); |
| 776 | if (TokenStream[CurIdx].is(K: tok::hashhash)) |
| 777 | ++CurIdx; |
| 778 | assert(!IsAtEnd() && "No token on the RHS of a paste operator!" ); |
| 779 | |
| 780 | // Get the RHS token. |
| 781 | const Token &RHS = TokenStream[CurIdx]; |
| 782 | |
| 783 | // Allocate space for the result token. This is guaranteed to be enough for |
| 784 | // the two tokens. |
| 785 | Buffer.resize(N: LHSTok.getLength() + RHS.getLength()); |
| 786 | |
| 787 | // Get the spelling of the LHS token in Buffer. |
| 788 | const char *BufPtr = &Buffer[0]; |
| 789 | bool Invalid = false; |
| 790 | unsigned LHSLen = PP.getSpelling(Tok: LHSTok, Buffer&: BufPtr, Invalid: &Invalid); |
| 791 | if (BufPtr != &Buffer[0]) // Really, we want the chars in Buffer! |
| 792 | memcpy(dest: &Buffer[0], src: BufPtr, n: LHSLen); |
| 793 | if (Invalid) |
| 794 | return true; |
| 795 | |
| 796 | BufPtr = Buffer.data() + LHSLen; |
| 797 | unsigned RHSLen = PP.getSpelling(Tok: RHS, Buffer&: BufPtr, Invalid: &Invalid); |
| 798 | if (Invalid) |
| 799 | return true; |
| 800 | if (RHSLen && BufPtr != &Buffer[LHSLen]) |
| 801 | // Really, we want the chars in Buffer! |
| 802 | memcpy(dest: &Buffer[LHSLen], src: BufPtr, n: RHSLen); |
| 803 | |
| 804 | // Trim excess space. |
| 805 | Buffer.resize(N: LHSLen+RHSLen); |
| 806 | |
| 807 | // Plop the pasted result (including the trailing newline and null) into a |
| 808 | // scratch buffer where we can lex it. |
| 809 | Token ResultTokTmp; |
| 810 | ResultTokTmp.startToken(); |
| 811 | |
| 812 | // Claim that the tmp token is a string_literal so that we can get the |
| 813 | // character pointer back from CreateString in getLiteralData(). |
| 814 | ResultTokTmp.setKind(tok::string_literal); |
| 815 | PP.CreateString(Str: Buffer, Tok&: ResultTokTmp); |
| 816 | SourceLocation ResultTokLoc = ResultTokTmp.getLocation(); |
| 817 | ResultTokStrPtr = ResultTokTmp.getLiteralData(); |
| 818 | |
| 819 | // Lex the resultant pasted token into Result. |
| 820 | Token Result; |
| 821 | |
| 822 | if (LHSTok.isAnyIdentifier() && RHS.isAnyIdentifier()) { |
| 823 | // Common paste case: identifier+identifier = identifier. Avoid creating |
| 824 | // a lexer and other overhead. |
| 825 | PP.IncrementPasteCounter(isFast: true); |
| 826 | Result.startToken(); |
| 827 | Result.setKind(tok::raw_identifier); |
| 828 | Result.setRawIdentifierData(ResultTokStrPtr); |
| 829 | Result.setLocation(ResultTokLoc); |
| 830 | Result.setLength(LHSLen+RHSLen); |
| 831 | } else { |
| 832 | PP.IncrementPasteCounter(isFast: false); |
| 833 | |
| 834 | assert(ResultTokLoc.isFileID() && |
| 835 | "Should be a raw location into scratch buffer" ); |
| 836 | SourceManager &SourceMgr = PP.getSourceManager(); |
| 837 | FileID LocFileID = SourceMgr.getFileID(SpellingLoc: ResultTokLoc); |
| 838 | |
| 839 | bool Invalid = false; |
| 840 | const char *ScratchBufStart |
| 841 | = SourceMgr.getBufferData(FID: LocFileID, Invalid: &Invalid).data(); |
| 842 | if (Invalid) |
| 843 | return false; |
| 844 | |
| 845 | // Make a lexer to lex this string from. Lex just this one token. |
| 846 | // Make a lexer object so that we lex and expand the paste result. |
| 847 | Lexer TL(SourceMgr.getLocForStartOfFile(FID: LocFileID), |
| 848 | PP.getLangOpts(), ScratchBufStart, |
| 849 | ResultTokStrPtr, ResultTokStrPtr+LHSLen+RHSLen); |
| 850 | |
| 851 | // Lex a token in raw mode. This way it won't look up identifiers |
| 852 | // automatically, lexing off the end will return an eof token, and |
| 853 | // warnings are disabled. This returns true if the result token is the |
| 854 | // entire buffer. |
| 855 | bool isInvalid = !TL.LexFromRawLexer(Result); |
| 856 | |
| 857 | // If we got an EOF token, we didn't form even ONE token. For example, we |
| 858 | // did "/ ## /" to get "//". |
| 859 | isInvalid |= Result.is(K: tok::eof); |
| 860 | |
| 861 | // If pasting the two tokens didn't form a full new token, this is an |
| 862 | // error. This occurs with "x ## +" and other stuff. Return with LHSTok |
| 863 | // unmodified and with RHS as the next token to lex. |
| 864 | if (isInvalid) { |
| 865 | // Explicitly convert the token location to have proper expansion |
| 866 | // information so that the user knows where it came from. |
| 867 | SourceManager &SM = PP.getSourceManager(); |
| 868 | SourceLocation Loc = |
| 869 | SM.createExpansionLoc(SpellingLoc: PasteOpLoc, ExpansionLocStart: ExpandLocStart, ExpansionLocEnd: ExpandLocEnd, Length: 2); |
| 870 | |
| 871 | // Test for the Microsoft extension of /##/ turning into // here on the |
| 872 | // error path. |
| 873 | if (PP.getLangOpts().MicrosoftExt && LHSTok.is(K: tok::slash) && |
| 874 | RHS.is(K: tok::slash)) { |
| 875 | HandleMicrosoftCommentPaste(Tok&: LHSTok, OpLoc: Loc); |
| 876 | return true; |
| 877 | } |
| 878 | |
| 879 | // Do not emit the error when preprocessing assembler code. |
| 880 | if (!PP.getLangOpts().AsmPreprocessor) { |
| 881 | // If we're in microsoft extensions mode, downgrade this from a hard |
| 882 | // error to an extension that defaults to an error. This allows |
| 883 | // disabling it. |
| 884 | PP.Diag(Loc, DiagID: PP.getLangOpts().MicrosoftExt ? diag::ext_pp_bad_paste_ms |
| 885 | : diag::err_pp_bad_paste) |
| 886 | << Buffer; |
| 887 | } |
| 888 | |
| 889 | // An error has occurred so exit loop. |
| 890 | break; |
| 891 | } |
| 892 | |
| 893 | // Turn ## into 'unknown' to avoid # ## # from looking like a paste |
| 894 | // operator. |
| 895 | if (Result.is(K: tok::hashhash)) |
| 896 | Result.setKind(tok::unknown); |
| 897 | } |
| 898 | |
| 899 | // Transfer properties of the LHS over the Result. |
| 900 | Result.setFlagValue(Flag: Token::StartOfLine , Val: LHSTok.isAtStartOfLine()); |
| 901 | Result.setFlagValue(Flag: Token::LeadingSpace, Val: LHSTok.hasLeadingSpace()); |
| 902 | |
| 903 | // Finally, replace LHS with the result, consume the RHS, and iterate. |
| 904 | ++CurIdx; |
| 905 | |
| 906 | // Set Token::HasUCN flag if LHS or RHS contains any UCNs. |
| 907 | HasUCNs = LHSTok.hasUCN() || RHS.hasUCN() || HasUCNs; |
| 908 | LHSTok = Result; |
| 909 | } while (!IsAtEnd() && TokenStream[CurIdx].is(K: tok::hashhash)); |
| 910 | |
| 911 | SourceLocation EndLoc = TokenStream[CurIdx - 1].getLocation(); |
| 912 | |
| 913 | // The token's current location indicate where the token was lexed from. We |
| 914 | // need this information to compute the spelling of the token, but any |
| 915 | // diagnostics for the expanded token should appear as if the token was |
| 916 | // expanded from the full ## expression. Pull this information together into |
| 917 | // a new SourceLocation that captures all of this. |
| 918 | SourceManager &SM = PP.getSourceManager(); |
| 919 | if (StartLoc.isFileID()) |
| 920 | StartLoc = getExpansionLocForMacroDefLoc(loc: StartLoc); |
| 921 | if (EndLoc.isFileID()) |
| 922 | EndLoc = getExpansionLocForMacroDefLoc(loc: EndLoc); |
| 923 | FileID MacroFID = SM.getFileID(SpellingLoc: MacroExpansionStart); |
| 924 | while (SM.getFileID(SpellingLoc: StartLoc) != MacroFID) |
| 925 | StartLoc = SM.getImmediateExpansionRange(Loc: StartLoc).getBegin(); |
| 926 | while (SM.getFileID(SpellingLoc: EndLoc) != MacroFID) |
| 927 | EndLoc = SM.getImmediateExpansionRange(Loc: EndLoc).getEnd(); |
| 928 | |
| 929 | LHSTok.setLocation(SM.createExpansionLoc(SpellingLoc: LHSTok.getLocation(), ExpansionLocStart: StartLoc, ExpansionLocEnd: EndLoc, |
| 930 | Length: LHSTok.getLength())); |
| 931 | |
| 932 | // Now that we got the result token, it will be subject to expansion. Since |
| 933 | // token pasting re-lexes the result token in raw mode, identifier information |
| 934 | // isn't looked up. As such, if the result is an identifier, look up id info. |
| 935 | if (LHSTok.is(K: tok::raw_identifier)) { |
| 936 | |
| 937 | // If there has any UNCs in concated token, we should mark this token |
| 938 | // with Token::HasUCN flag, then LookUpIdentifierInfo will expand UCNs in |
| 939 | // token. |
| 940 | if (HasUCNs) |
| 941 | LHSTok.setFlag(Token::HasUCN); |
| 942 | |
| 943 | // Look up the identifier info for the token. We disabled identifier lookup |
| 944 | // by saying we're skipping contents, so we need to do this manually. |
| 945 | PP.LookUpIdentifierInfo(Identifier&: LHSTok); |
| 946 | } |
| 947 | return false; |
| 948 | } |
| 949 | |
| 950 | /// isNextTokenLParen - If the next token lexed will pop this macro off the |
| 951 | /// expansion stack, return std::nullopt, otherwise return the next unexpanded |
| 952 | /// token. |
| 953 | std::optional<Token> TokenLexer::peekNextPPToken() const { |
| 954 | // Out of tokens? |
| 955 | if (isAtEnd()) |
| 956 | return std::nullopt; |
| 957 | return Tokens[CurTokenIdx]; |
| 958 | } |
| 959 | |
| 960 | /// isParsingPreprocessorDirective - Return true if we are in the middle of a |
| 961 | /// preprocessor directive. |
| 962 | bool TokenLexer::isParsingPreprocessorDirective() const { |
| 963 | return Tokens[NumTokens-1].is(K: tok::eod) && !isAtEnd(); |
| 964 | } |
| 965 | |
| 966 | /// setLexingCXXModuleDirective - This is set to true if this TokenLexer is |
| 967 | /// created when handling C++ module directive. |
| 968 | void TokenLexer::setLexingCXXModuleDirective(bool Val) { |
| 969 | LexingCXXModuleDirective = Val; |
| 970 | } |
| 971 | |
| 972 | /// isLexingCXXModuleDirective - Return true if we are lexing a C++ module or |
| 973 | /// import directive. |
| 974 | bool TokenLexer::isLexingCXXModuleDirective() const { |
| 975 | return LexingCXXModuleDirective; |
| 976 | } |
| 977 | |
| 978 | /// HandleMicrosoftCommentPaste - In microsoft compatibility mode, /##/ pastes |
| 979 | /// together to form a comment that comments out everything in the current |
| 980 | /// macro, other active macros, and anything left on the current physical |
| 981 | /// source line of the expanded buffer. Handle this by returning the |
| 982 | /// first token on the next line. |
| 983 | void TokenLexer::HandleMicrosoftCommentPaste(Token &Tok, SourceLocation OpLoc) { |
| 984 | PP.Diag(Loc: OpLoc, DiagID: diag::ext_comment_paste_microsoft); |
| 985 | |
| 986 | // We 'comment out' the rest of this macro by just ignoring the rest of the |
| 987 | // tokens that have not been lexed yet, if any. |
| 988 | |
| 989 | // Since this must be a macro, mark the macro enabled now that it is no longer |
| 990 | // being expanded. |
| 991 | assert(Macro && "Token streams can't paste comments" ); |
| 992 | Macro->EnableMacro(); |
| 993 | |
| 994 | PP.HandleMicrosoftCommentPaste(Tok); |
| 995 | } |
| 996 | |
| 997 | /// If \arg loc is a file ID and points inside the current macro |
| 998 | /// definition, returns the appropriate source location pointing at the |
| 999 | /// macro expansion source location entry, otherwise it returns an invalid |
| 1000 | /// SourceLocation. |
| 1001 | SourceLocation |
| 1002 | TokenLexer::getExpansionLocForMacroDefLoc(SourceLocation loc) const { |
| 1003 | assert(ExpandLocStart.isValid() && MacroExpansionStart.isValid() && |
| 1004 | "Not appropriate for token streams" ); |
| 1005 | assert(loc.isValid() && loc.isFileID()); |
| 1006 | |
| 1007 | SourceManager &SM = PP.getSourceManager(); |
| 1008 | assert(SM.isInSLocAddrSpace(loc, MacroDefStart, MacroDefLength) && |
| 1009 | "Expected loc to come from the macro definition" ); |
| 1010 | |
| 1011 | SourceLocation::UIntTy relativeOffset = 0; |
| 1012 | SM.isInSLocAddrSpace(Loc: loc, Start: MacroDefStart, Length: MacroDefLength, RelativeOffset: &relativeOffset); |
| 1013 | return MacroExpansionStart.getLocWithOffset(Offset: relativeOffset); |
| 1014 | } |
| 1015 | |
| 1016 | /// Finds the tokens that are consecutive (from the same FileID) |
| 1017 | /// creates a single SLocEntry, and assigns SourceLocations to each token that |
| 1018 | /// point to that SLocEntry. e.g for |
| 1019 | /// assert(foo == bar); |
| 1020 | /// There will be a single SLocEntry for the "foo == bar" chunk and locations |
| 1021 | /// for the 'foo', '==', 'bar' tokens will point inside that chunk. |
| 1022 | /// |
| 1023 | /// \arg begin_tokens will be updated to a position past all the found |
| 1024 | /// consecutive tokens. |
| 1025 | static void updateConsecutiveMacroArgTokens(SourceManager &SM, |
| 1026 | SourceLocation ExpandLoc, |
| 1027 | Token *&begin_tokens, |
| 1028 | Token * end_tokens) { |
| 1029 | assert(begin_tokens + 1 < end_tokens); |
| 1030 | SourceLocation BeginLoc = begin_tokens->getLocation(); |
| 1031 | llvm::MutableArrayRef<Token> All(begin_tokens, end_tokens); |
| 1032 | llvm::MutableArrayRef<Token> Partition; |
| 1033 | |
| 1034 | auto NearLast = [&, Last = BeginLoc](SourceLocation Loc) mutable { |
| 1035 | // The maximum distance between two consecutive tokens in a partition. |
| 1036 | // This is an important trick to avoid using too much SourceLocation address |
| 1037 | // space! |
| 1038 | static constexpr SourceLocation::IntTy MaxDistance = 50; |
| 1039 | auto Distance = Loc.getRawEncoding() - Last.getRawEncoding(); |
| 1040 | Last = Loc; |
| 1041 | return Distance <= MaxDistance; |
| 1042 | }; |
| 1043 | |
| 1044 | // Partition the tokens by their FileID. |
| 1045 | // This is a hot function, and calling getFileID can be expensive, the |
| 1046 | // implementation is optimized by reducing the number of getFileID. |
| 1047 | if (BeginLoc.isFileID()) { |
| 1048 | // Consecutive tokens not written in macros must be from the same file. |
| 1049 | // (Neither #include nor eof can occur inside a macro argument.) |
| 1050 | Partition = All.take_while(Pred: [&](const Token &T) { |
| 1051 | return T.getLocation().isFileID() && NearLast(T.getLocation()); |
| 1052 | }); |
| 1053 | } else { |
| 1054 | // Call getFileID once to calculate the bounds, and use the cheaper |
| 1055 | // sourcelocation-against-bounds comparison. |
| 1056 | FileID BeginFID = SM.getFileID(SpellingLoc: BeginLoc); |
| 1057 | SourceLocation Limit = |
| 1058 | SM.getComposedLoc(FID: BeginFID, Offset: SM.getFileIDSize(FID: BeginFID)); |
| 1059 | Partition = All.take_while(Pred: [&](const Token &T) { |
| 1060 | // NOTE: the Limit is included! The lexer recovery only ever inserts a |
| 1061 | // single token past the end of the FileID, specifically the ) when a |
| 1062 | // macro-arg containing a comma should be guarded by parentheses. |
| 1063 | // |
| 1064 | // It is safe to include the Limit here because SourceManager allocates |
| 1065 | // FileSize + 1 for each SLocEntry. |
| 1066 | // |
| 1067 | // See https://github.com/llvm/llvm-project/issues/60722. |
| 1068 | return T.getLocation() >= BeginLoc && T.getLocation() <= Limit |
| 1069 | && NearLast(T.getLocation()); |
| 1070 | }); |
| 1071 | } |
| 1072 | assert(!Partition.empty()); |
| 1073 | |
| 1074 | // For the consecutive tokens, find the length of the SLocEntry to contain |
| 1075 | // all of them. |
| 1076 | SourceLocation::UIntTy FullLength = |
| 1077 | Partition.back().getEndLoc().getRawEncoding() - |
| 1078 | Partition.front().getLocation().getRawEncoding(); |
| 1079 | // Create a macro expansion SLocEntry that will "contain" all of the tokens. |
| 1080 | SourceLocation Expansion = |
| 1081 | SM.createMacroArgExpansionLoc(SpellingLoc: BeginLoc, ExpansionLoc: ExpandLoc, Length: FullLength); |
| 1082 | |
| 1083 | #ifdef EXPENSIVE_CHECKS |
| 1084 | assert(llvm::all_of(Partition.drop_front(), |
| 1085 | [&SM, ID = SM.getFileID(Partition.front().getLocation())]( |
| 1086 | const Token &T) { |
| 1087 | return ID == SM.getFileID(T.getLocation()); |
| 1088 | }) && |
| 1089 | "Must have the same FIleID!" ); |
| 1090 | #endif |
| 1091 | // Change the location of the tokens from the spelling location to the new |
| 1092 | // expanded location. |
| 1093 | for (Token& T : Partition) { |
| 1094 | SourceLocation::IntTy RelativeOffset = |
| 1095 | T.getLocation().getRawEncoding() - BeginLoc.getRawEncoding(); |
| 1096 | T.setLocation(Expansion.getLocWithOffset(Offset: RelativeOffset)); |
| 1097 | } |
| 1098 | begin_tokens = &Partition.back() + 1; |
| 1099 | } |
| 1100 | |
| 1101 | /// Creates SLocEntries and updates the locations of macro argument |
| 1102 | /// tokens to their new expanded locations. |
| 1103 | /// |
| 1104 | /// \param ArgIdSpellLoc the location of the macro argument id inside the macro |
| 1105 | /// definition. |
| 1106 | void TokenLexer::updateLocForMacroArgTokens(SourceLocation ArgIdSpellLoc, |
| 1107 | Token *begin_tokens, |
| 1108 | Token *end_tokens) { |
| 1109 | SourceManager &SM = PP.getSourceManager(); |
| 1110 | |
| 1111 | SourceLocation ExpandLoc = |
| 1112 | getExpansionLocForMacroDefLoc(loc: ArgIdSpellLoc); |
| 1113 | |
| 1114 | while (begin_tokens < end_tokens) { |
| 1115 | // If there's only one token just create a SLocEntry for it. |
| 1116 | if (end_tokens - begin_tokens == 1) { |
| 1117 | Token &Tok = *begin_tokens; |
| 1118 | Tok.setLocation(SM.createMacroArgExpansionLoc(SpellingLoc: Tok.getLocation(), |
| 1119 | ExpansionLoc: ExpandLoc, |
| 1120 | Length: Tok.getLength())); |
| 1121 | return; |
| 1122 | } |
| 1123 | |
| 1124 | updateConsecutiveMacroArgTokens(SM, ExpandLoc, begin_tokens, end_tokens); |
| 1125 | } |
| 1126 | } |
| 1127 | |
| 1128 | void TokenLexer::PropagateLineStartLeadingSpaceInfo(Token &Result) { |
| 1129 | AtStartOfLine = Result.isAtStartOfLine(); |
| 1130 | HasLeadingSpace = Result.hasLeadingSpace(); |
| 1131 | } |
| 1132 | |