TokenLexer.cpp revision 360660
1158115Sume//===- TokenLexer.cpp - Lex from a token stream ---------------------------===// 2158115Sume// 3158115Sume// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4158115Sume// See https://llvm.org/LICENSE.txt for license information. 5158115Sume// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6158115Sume// 7158115Sume//===----------------------------------------------------------------------===// 8158115Sume// 9158115Sume// This file implements the TokenLexer interface. 10158115Sume// 11158115Sume//===----------------------------------------------------------------------===// 12158115Sume 13158115Sume#include "clang/Lex/TokenLexer.h" 14158115Sume#include "clang/Basic/Diagnostic.h" 15158115Sume#include "clang/Basic/IdentifierTable.h" 16158115Sume#include "clang/Basic/LangOptions.h" 17158115Sume#include "clang/Basic/SourceLocation.h" 18158115Sume#include "clang/Basic/SourceManager.h" 19158115Sume#include "clang/Basic/TokenKinds.h" 20158115Sume#include "clang/Lex/LexDiagnostic.h" 21158115Sume#include "clang/Lex/Lexer.h" 22158115Sume#include "clang/Lex/MacroArgs.h" 23158115Sume#include "clang/Lex/MacroInfo.h" 24158115Sume#include "clang/Lex/Preprocessor.h" 25158115Sume#include "clang/Lex/Token.h" 26158115Sume#include "clang/Lex/VariadicMacroSupport.h" 27158115Sume#include "llvm/ADT/ArrayRef.h" 28158115Sume#include "llvm/ADT/SmallString.h" 29158115Sume#include "llvm/ADT/SmallVector.h" 30158115Sume#include "llvm/ADT/iterator_range.h" 31158115Sume#include <cassert> 32158115Sume#include <cstring> 33158115Sume 34158115Sumeusing namespace clang; 35158115Sume 36158115Sume/// Create a TokenLexer for the specified macro with the specified actual 37158115Sume/// arguments. Note that this ctor takes ownership of the ActualArgs pointer. 38158115Sumevoid TokenLexer::Init(Token &Tok, SourceLocation ELEnd, MacroInfo *MI, 39158115Sume MacroArgs *Actuals) { 40158115Sume // If the client is reusing a TokenLexer, make sure to free any memory 41158115Sume // associated with it. 42158115Sume destroy(); 43158115Sume 44158115Sume Macro = MI; 45158115Sume ActualArgs = Actuals; 46158115Sume CurTokenIdx = 0; 47158115Sume 48158115Sume ExpandLocStart = Tok.getLocation(); 49158115Sume ExpandLocEnd = ELEnd; 50158115Sume AtStartOfLine = Tok.isAtStartOfLine(); 51158115Sume HasLeadingSpace = Tok.hasLeadingSpace(); 52158115Sume NextTokGetsSpace = false; 53158115Sume Tokens = &*Macro->tokens_begin(); 54158115Sume OwnsTokens = false; 55158115Sume DisableMacroExpansion = false; 56158115Sume IsReinject = false; 57158115Sume NumTokens = Macro->tokens_end()-Macro->tokens_begin(); 58158115Sume MacroExpansionStart = SourceLocation(); 59158115Sume 60158115Sume SourceManager &SM = PP.getSourceManager(); 61158115Sume MacroStartSLocOffset = SM.getNextLocalOffset(); 62158115Sume 63158115Sume if (NumTokens > 0) { 64158115Sume assert(Tokens[0].getLocation().isValid()); 65158115Sume assert((Tokens[0].getLocation().isFileID() || Tokens[0].is(tok::comment)) && 66158115Sume "Macro defined in macro?"); 67158115Sume assert(ExpandLocStart.isValid()); 68158115Sume 69158115Sume // Reserve a source location entry chunk for the length of the macro 70158115Sume // definition. Tokens that get lexed directly from the definition will 71158115Sume // have their locations pointing inside this chunk. This is to avoid 72158115Sume // creating separate source location entries for each token. 73158115Sume MacroDefStart = SM.getExpansionLoc(Tokens[0].getLocation()); 74158115Sume MacroDefLength = Macro->getDefinitionLength(SM); 75158115Sume MacroExpansionStart = SM.createExpansionLoc(MacroDefStart, 76158115Sume ExpandLocStart, 77158115Sume ExpandLocEnd, 78158115Sume MacroDefLength); 79158115Sume } 80158115Sume 81158115Sume // If this is a function-like macro, expand the arguments and change 82158115Sume // Tokens to point to the expanded tokens. 83158115Sume if (Macro->isFunctionLike() && Macro->getNumParams()) 84158115Sume ExpandFunctionArguments(); 85158115Sume 86158115Sume // Mark the macro as currently disabled, so that it is not recursively 87158115Sume // expanded. The macro must be disabled only after argument pre-expansion of 88158115Sume // function-like macro arguments occurs. 89158115Sume Macro->DisableMacro(); 90158115Sume} 91158115Sume 92158115Sume/// Create a TokenLexer for the specified token stream. This does not 93158115Sume/// take ownership of the specified token vector. 94158115Sumevoid TokenLexer::Init(const Token *TokArray, unsigned NumToks, 95158115Sume bool disableMacroExpansion, bool ownsTokens, 96158115Sume bool isReinject) { 97158115Sume assert(!isReinject || disableMacroExpansion); 98158115Sume // If the client is reusing a TokenLexer, make sure to free any memory 99158115Sume // associated with it. 100158115Sume destroy(); 101158115Sume 102158115Sume Macro = nullptr; 103158115Sume ActualArgs = nullptr; 104158115Sume Tokens = TokArray; 105158115Sume OwnsTokens = ownsTokens; 106158115Sume DisableMacroExpansion = disableMacroExpansion; 107158115Sume IsReinject = isReinject; 108158115Sume NumTokens = NumToks; 109158115Sume CurTokenIdx = 0; 110158115Sume ExpandLocStart = ExpandLocEnd = SourceLocation(); 111158115Sume AtStartOfLine = false; 112158115Sume HasLeadingSpace = false; 113158115Sume NextTokGetsSpace = false; 114158115Sume MacroExpansionStart = SourceLocation(); 115158115Sume 116158115Sume // Set HasLeadingSpace/AtStartOfLine so that the first token will be 117158115Sume // returned unmodified. 118158115Sume if (NumToks != 0) { 119158115Sume AtStartOfLine = TokArray[0].isAtStartOfLine(); 120158115Sume HasLeadingSpace = TokArray[0].hasLeadingSpace(); 121158115Sume } 122158115Sume} 123158115Sume 124158115Sumevoid TokenLexer::destroy() { 125158115Sume // If this was a function-like macro that actually uses its arguments, delete 126158115Sume // the expanded tokens. 127158115Sume if (OwnsTokens) { 128158115Sume delete [] Tokens; 129158115Sume Tokens = nullptr; 130158115Sume OwnsTokens = false; 131158115Sume } 132158115Sume 133158115Sume // TokenLexer owns its formal arguments. 134158115Sume if (ActualArgs) ActualArgs->destroy(PP); 135158115Sume} 136158115Sume 137158115Sumebool TokenLexer::MaybeRemoveCommaBeforeVaArgs( 138158115Sume SmallVectorImpl<Token> &ResultToks, bool HasPasteOperator, MacroInfo *Macro, 139158115Sume unsigned MacroArgNo, Preprocessor &PP) { 140158115Sume // Is the macro argument __VA_ARGS__? 141158115Sume if (!Macro->isVariadic() || MacroArgNo != Macro->getNumParams()-1) 142158115Sume return false; 143158115Sume 144158115Sume // In Microsoft-compatibility mode, a comma is removed in the expansion 145158115Sume // of " ... , __VA_ARGS__ " if __VA_ARGS__ is empty. This extension is 146158115Sume // not supported by gcc. 147158115Sume if (!HasPasteOperator && !PP.getLangOpts().MSVCCompat) 148158115Sume return false; 149158115Sume 150158115Sume // GCC removes the comma in the expansion of " ... , ## __VA_ARGS__ " if 151158115Sume // __VA_ARGS__ is empty, but not in strict C99 mode where there are no 152158115Sume // named arguments, where it remains. In all other modes, including C99 153158115Sume // with GNU extensions, it is removed regardless of named arguments. 154158115Sume // Microsoft also appears to support this extension, unofficially. 155158115Sume if (PP.getLangOpts().C99 && !PP.getLangOpts().GNUMode 156158115Sume && Macro->getNumParams() < 2) 157158115Sume return false; 158158115Sume 159158115Sume // Is a comma available to be removed? 160158115Sume if (ResultToks.empty() || !ResultToks.back().is(tok::comma)) 161158115Sume return false; 162158115Sume 163158115Sume // Issue an extension diagnostic for the paste operator. 164158115Sume if (HasPasteOperator) 165158115Sume PP.Diag(ResultToks.back().getLocation(), diag::ext_paste_comma); 166158115Sume 167158115Sume // Remove the comma. 168158115Sume ResultToks.pop_back(); 169158115Sume 170158115Sume if (!ResultToks.empty()) { 171158115Sume // If the comma was right after another paste (e.g. "X##,##__VA_ARGS__"), 172158115Sume // then removal of the comma should produce a placemarker token (in C99 173158115Sume // terms) which we model by popping off the previous ##, giving us a plain 174158115Sume // "X" when __VA_ARGS__ is empty. 175158115Sume if (ResultToks.back().is(tok::hashhash)) 176158115Sume ResultToks.pop_back(); 177158115Sume 178158115Sume // Remember that this comma was elided. 179158115Sume ResultToks.back().setFlag(Token::CommaAfterElided); 180158115Sume } 181158115Sume 182158115Sume // Never add a space, even if the comma, ##, or arg had a space. 183158115Sume NextTokGetsSpace = false; 184158115Sume return true; 185158115Sume} 186158115Sume 187158115Sumevoid TokenLexer::stringifyVAOPTContents( 188158115Sume SmallVectorImpl<Token> &ResultToks, const VAOptExpansionContext &VCtx, 189158115Sume const SourceLocation VAOPTClosingParenLoc) { 190158115Sume const int NumToksPriorToVAOpt = VCtx.getNumberOfTokensPriorToVAOpt(); 191158115Sume const unsigned int NumVAOptTokens = ResultToks.size() - NumToksPriorToVAOpt; 192158115Sume Token *const VAOPTTokens = 193158115Sume NumVAOptTokens ? &ResultToks[NumToksPriorToVAOpt] : nullptr; 194158115Sume 195158115Sume SmallVector<Token, 64> ConcatenatedVAOPTResultToks; 196158115Sume // FIXME: Should we keep track within VCtx that we did or didnot 197158115Sume // encounter pasting - and only then perform this loop. 198158115Sume 199158115Sume // Perform token pasting (concatenation) prior to stringization. 200158115Sume for (unsigned int CurTokenIdx = 0; CurTokenIdx != NumVAOptTokens; 201158115Sume ++CurTokenIdx) { 202158115Sume if (VAOPTTokens[CurTokenIdx].is(tok::hashhash)) { 203158115Sume assert(CurTokenIdx != 0 && 204158115Sume "Can not have __VAOPT__ contents begin with a ##"); 205158115Sume Token &LHS = VAOPTTokens[CurTokenIdx - 1]; 206158115Sume pasteTokens(LHS, llvm::makeArrayRef(VAOPTTokens, NumVAOptTokens), 207158115Sume CurTokenIdx); 208158115Sume // Replace the token prior to the first ## in this iteration. 209158115Sume ConcatenatedVAOPTResultToks.back() = LHS; 210158115Sume if (CurTokenIdx == NumVAOptTokens) 211158115Sume break; 212158115Sume } 213158115Sume ConcatenatedVAOPTResultToks.push_back(VAOPTTokens[CurTokenIdx]); 214158115Sume } 215158115Sume 216158115Sume ConcatenatedVAOPTResultToks.push_back(VCtx.getEOFTok()); 217158115Sume // Get the SourceLocation that represents the start location within 218158115Sume // the macro definition that marks where this string is substituted 219158115Sume // into: i.e. the __VA_OPT__ and the ')' within the spelling of the 220158115Sume // macro definition, and use it to indicate that the stringified token 221158115Sume // was generated from that location. 222158115Sume const SourceLocation ExpansionLocStartWithinMacro = 223158115Sume getExpansionLocForMacroDefLoc(VCtx.getVAOptLoc()); 224158115Sume const SourceLocation ExpansionLocEndWithinMacro = 225158115Sume getExpansionLocForMacroDefLoc(VAOPTClosingParenLoc); 226158115Sume 227158115Sume Token StringifiedVAOPT = MacroArgs::StringifyArgument( 228158115Sume &ConcatenatedVAOPTResultToks[0], PP, VCtx.hasCharifyBefore() /*Charify*/, 229158115Sume ExpansionLocStartWithinMacro, ExpansionLocEndWithinMacro); 230158115Sume 231158115Sume if (VCtx.getLeadingSpaceForStringifiedToken()) 232158115Sume StringifiedVAOPT.setFlag(Token::LeadingSpace); 233158115Sume 234158115Sume StringifiedVAOPT.setFlag(Token::StringifiedInMacro); 235158115Sume // Resize (shrink) the token stream to just capture this stringified token. 236158115Sume ResultToks.resize(NumToksPriorToVAOpt + 1); 237158115Sume ResultToks.back() = StringifiedVAOPT; 238158115Sume} 239158115Sume 240158115Sume/// Expand the arguments of a function-like macro so that we can quickly 241158115Sume/// return preexpanded tokens from Tokens. 242158115Sumevoid TokenLexer::ExpandFunctionArguments() { 243158115Sume SmallVector<Token, 128> ResultToks; 244158115Sume 245158115Sume // Loop through 'Tokens', expanding them into ResultToks. Keep 246158115Sume // track of whether we change anything. If not, no need to keep them. If so, 247158115Sume // we install the newly expanded sequence as the new 'Tokens' list. 248158115Sume bool MadeChange = false; 249158115Sume 250158115Sume Optional<bool> CalledWithVariadicArguments; 251158115Sume 252158115Sume VAOptExpansionContext VCtx(PP); 253158115Sume 254158115Sume for (unsigned I = 0, E = NumTokens; I != E; ++I) { 255158115Sume const Token &CurTok = Tokens[I]; 256158115Sume // We don't want a space for the next token after a paste 257158115Sume // operator. In valid code, the token will get smooshed onto the 258158115Sume // preceding one anyway. In assembler-with-cpp mode, invalid 259158115Sume // pastes are allowed through: in this case, we do not want the 260158115Sume // extra whitespace to be added. For example, we want ". ## foo" 261158115Sume // -> ".foo" not ". foo". 262158115Sume if (I != 0 && !Tokens[I-1].is(tok::hashhash) && CurTok.hasLeadingSpace()) 263158115Sume NextTokGetsSpace = true; 264158115Sume 265158115Sume if (VCtx.isVAOptToken(CurTok)) { 266158115Sume MadeChange = true; 267158115Sume assert(Tokens[I + 1].is(tok::l_paren) && 268158115Sume "__VA_OPT__ must be followed by '('"); 269158115Sume 270158115Sume ++I; // Skip the l_paren 271158115Sume VCtx.sawVAOptFollowedByOpeningParens(CurTok.getLocation(), 272158115Sume ResultToks.size()); 273158115Sume 274158115Sume continue; 275158115Sume } 276158115Sume 277158115Sume // We have entered into the __VA_OPT__ context, so handle tokens 278158115Sume // appropriately. 279158115Sume if (VCtx.isInVAOpt()) { 280158115Sume // If we are about to process a token that is either an argument to 281158115Sume // __VA_OPT__ or its closing rparen, then: 282158115Sume // 1) If the token is the closing rparen that exits us out of __VA_OPT__, 283158115Sume // perform any necessary stringification or placemarker processing, 284158115Sume // and/or skip to the next token. 285158115Sume // 2) else if macro was invoked without variadic arguments skip this 286158115Sume // token. 287158115Sume // 3) else (macro was invoked with variadic arguments) process the token 288158115Sume // normally. 289158115Sume 290158115Sume if (Tokens[I].is(tok::l_paren)) 291158115Sume VCtx.sawOpeningParen(Tokens[I].getLocation()); 292158115Sume // Continue skipping tokens within __VA_OPT__ if the macro was not 293158115Sume // called with variadic arguments, else let the rest of the loop handle 294158115Sume // this token. Note sawClosingParen() returns true only if the r_paren matches 295158115Sume // the closing r_paren of the __VA_OPT__. 296158115Sume if (!Tokens[I].is(tok::r_paren) || !VCtx.sawClosingParen()) { 297158115Sume // Lazily expand __VA_ARGS__ when we see the first __VA_OPT__. 298158115Sume if (!CalledWithVariadicArguments.hasValue()) { 299158115Sume CalledWithVariadicArguments = 300158115Sume ActualArgs->invokedWithVariadicArgument(Macro, PP); 301158115Sume } 302158115Sume if (!*CalledWithVariadicArguments) { 303158115Sume // Skip this token. 304158115Sume continue; 305158115Sume } 306158115Sume // ... else the macro was called with variadic arguments, and we do not 307158115Sume // have a closing rparen - so process this token normally. 308158115Sume } else { 309158115Sume // Current token is the closing r_paren which marks the end of the 310158115Sume // __VA_OPT__ invocation, so handle any place-marker pasting (if 311158115Sume // empty) by removing hashhash either before (if exists) or after. And 312158115Sume // also stringify the entire contents if VAOPT was preceded by a hash, 313158115Sume // but do so only after any token concatenation that needs to occur 314158115Sume // within the contents of VAOPT. 315158115Sume 316158115Sume if (VCtx.hasStringifyOrCharifyBefore()) { 317158115Sume // Replace all the tokens just added from within VAOPT into a single 318158115Sume // stringified token. This requires token-pasting to eagerly occur 319158115Sume // within these tokens. If either the contents of VAOPT were empty 320158115Sume // or the macro wasn't called with any variadic arguments, the result 321158115Sume // is a token that represents an empty string. 322158115Sume stringifyVAOPTContents(ResultToks, VCtx, 323158115Sume /*ClosingParenLoc*/ Tokens[I].getLocation()); 324158115Sume 325158115Sume } else if (/*No tokens within VAOPT*/ 326158115Sume ResultToks.size() == VCtx.getNumberOfTokensPriorToVAOpt()) { 327158115Sume // Treat VAOPT as a placemarker token. Eat either the '##' before the 328158115Sume // RHS/VAOPT (if one exists, suggesting that the LHS (if any) to that 329158115Sume // hashhash was not a placemarker) or the '##' 330158115Sume // after VAOPT, but not both. 331158115Sume 332158115Sume if (ResultToks.size() && ResultToks.back().is(tok::hashhash)) { 333158115Sume ResultToks.pop_back(); 334158115Sume } else if ((I + 1 != E) && Tokens[I + 1].is(tok::hashhash)) { 335158115Sume ++I; // Skip the following hashhash. 336158115Sume } 337158115Sume } else { 338158115Sume // If there's a ## before the __VA_OPT__, we might have discovered 339158115Sume // that the __VA_OPT__ begins with a placeholder. We delay action on 340158115Sume // that to now to avoid messing up our stashed count of tokens before 341158115Sume // __VA_OPT__. 342158115Sume if (VCtx.beginsWithPlaceholder()) { 343158115Sume assert(VCtx.getNumberOfTokensPriorToVAOpt() > 0 && 344158115Sume ResultToks.size() >= VCtx.getNumberOfTokensPriorToVAOpt() && 345158115Sume ResultToks[VCtx.getNumberOfTokensPriorToVAOpt() - 1].is( 346158115Sume tok::hashhash) && 347158115Sume "no token paste before __VA_OPT__"); 348158115Sume ResultToks.erase(ResultToks.begin() + 349158115Sume VCtx.getNumberOfTokensPriorToVAOpt() - 1); 350158115Sume } 351158115Sume // If the expansion of __VA_OPT__ ends with a placeholder, eat any 352158115Sume // following '##' token. 353158115Sume if (VCtx.endsWithPlaceholder() && I + 1 != E && 354158115Sume Tokens[I + 1].is(tok::hashhash)) { 355158115Sume ++I; 356158115Sume } 357158115Sume } 358158115Sume VCtx.reset(); 359158115Sume // We processed __VA_OPT__'s closing paren (and the exit out of 360158115Sume // __VA_OPT__), so skip to the next token. 361158115Sume continue; 362158115Sume } 363158115Sume } 364158115Sume 365158115Sume // If we found the stringify operator, get the argument stringified. The 366158115Sume // preprocessor already verified that the following token is a macro 367158115Sume // parameter or __VA_OPT__ when the #define was lexed. 368158115Sume 369158115Sume if (CurTok.isOneOf(tok::hash, tok::hashat)) { 370158115Sume int ArgNo = Macro->getParameterNum(Tokens[I+1].getIdentifierInfo()); 371158115Sume assert((ArgNo != -1 || VCtx.isVAOptToken(Tokens[I + 1])) && 372158115Sume "Token following # is not an argument or __VA_OPT__!"); 373158115Sume 374158115Sume if (ArgNo == -1) { 375158115Sume // Handle the __VA_OPT__ case. 376158115Sume VCtx.sawHashOrHashAtBefore(NextTokGetsSpace, 377158115Sume CurTok.is(tok::hashat)); 378158115Sume continue; 379158115Sume } 380158115Sume // Else handle the simple argument case. 381158115Sume SourceLocation ExpansionLocStart = 382158115Sume getExpansionLocForMacroDefLoc(CurTok.getLocation()); 383158115Sume SourceLocation ExpansionLocEnd = 384158115Sume getExpansionLocForMacroDefLoc(Tokens[I+1].getLocation()); 385158115Sume 386158115Sume Token Res; 387158115Sume if (CurTok.is(tok::hash)) // Stringify 388158115Sume Res = ActualArgs->getStringifiedArgument(ArgNo, PP, 389158115Sume ExpansionLocStart, 390158115Sume ExpansionLocEnd); 391158115Sume else { 392158115Sume // 'charify': don't bother caching these. 393158115Sume Res = MacroArgs::StringifyArgument(ActualArgs->getUnexpArgument(ArgNo), 394158115Sume PP, true, 395158115Sume ExpansionLocStart, 396158115Sume ExpansionLocEnd); 397158115Sume } 398158115Sume Res.setFlag(Token::StringifiedInMacro); 399158115Sume 400158115Sume // The stringified/charified string leading space flag gets set to match 401158115Sume // the #/#@ operator. 402158115Sume if (NextTokGetsSpace) 403158115Sume Res.setFlag(Token::LeadingSpace); 404158115Sume 405158115Sume ResultToks.push_back(Res); 406158115Sume MadeChange = true; 407158115Sume ++I; // Skip arg name. 408158115Sume NextTokGetsSpace = false; 409158115Sume continue; 410158115Sume } 411158115Sume 412158115Sume // Find out if there is a paste (##) operator before or after the token. 413158115Sume bool NonEmptyPasteBefore = 414158115Sume !ResultToks.empty() && ResultToks.back().is(tok::hashhash); 415158115Sume bool PasteBefore = I != 0 && Tokens[I-1].is(tok::hashhash); 416158115Sume bool PasteAfter = I+1 != E && Tokens[I+1].is(tok::hashhash); 417158115Sume bool RParenAfter = I+1 != E && Tokens[I+1].is(tok::r_paren); 418158115Sume 419158115Sume assert((!NonEmptyPasteBefore || PasteBefore || VCtx.isInVAOpt()) && 420158115Sume "unexpected ## in ResultToks"); 421158115Sume 422158115Sume // Otherwise, if this is not an argument token, just add the token to the 423158115Sume // output buffer. 424158115Sume IdentifierInfo *II = CurTok.getIdentifierInfo(); 425158115Sume int ArgNo = II ? Macro->getParameterNum(II) : -1; 426158115Sume if (ArgNo == -1) { 427158115Sume // This isn't an argument, just add it. 428158115Sume ResultToks.push_back(CurTok); 429158115Sume 430158115Sume if (NextTokGetsSpace) { 431158115Sume ResultToks.back().setFlag(Token::LeadingSpace); 432158115Sume NextTokGetsSpace = false; 433158115Sume } else if (PasteBefore && !NonEmptyPasteBefore) 434158115Sume ResultToks.back().clearFlag(Token::LeadingSpace); 435158115Sume 436158115Sume continue; 437158115Sume } 438158115Sume 439158115Sume // An argument is expanded somehow, the result is different than the 440158115Sume // input. 441158115Sume MadeChange = true; 442158115Sume 443158115Sume // Otherwise, this is a use of the argument. 444158115Sume 445158115Sume // In Microsoft mode, remove the comma before __VA_ARGS__ to ensure there 446158115Sume // are no trailing commas if __VA_ARGS__ is empty. 447158115Sume if (!PasteBefore && ActualArgs->isVarargsElidedUse() && 448158115Sume MaybeRemoveCommaBeforeVaArgs(ResultToks, 449158115Sume /*HasPasteOperator=*/false, 450158115Sume Macro, ArgNo, PP)) 451158115Sume continue; 452158115Sume 453158115Sume // If it is not the LHS/RHS of a ## operator, we must pre-expand the 454158115Sume // argument and substitute the expanded tokens into the result. This is 455158115Sume // C99 6.10.3.1p1. 456158115Sume if (!PasteBefore && !PasteAfter) { 457158115Sume const Token *ResultArgToks; 458158115Sume 459158115Sume // Only preexpand the argument if it could possibly need it. This 460158115Sume // avoids some work in common cases. 461158115Sume const Token *ArgTok = ActualArgs->getUnexpArgument(ArgNo); 462158115Sume if (ActualArgs->ArgNeedsPreexpansion(ArgTok, PP)) 463158115Sume ResultArgToks = &ActualArgs->getPreExpArgument(ArgNo, PP)[0]; 464158115Sume else 465158115Sume ResultArgToks = ArgTok; // Use non-preexpanded tokens. 466158115Sume 467158115Sume // If the arg token expanded into anything, append it. 468158115Sume if (ResultArgToks->isNot(tok::eof)) { 469158115Sume size_t FirstResult = ResultToks.size(); 470158115Sume unsigned NumToks = MacroArgs::getArgLength(ResultArgToks); 471158115Sume ResultToks.append(ResultArgToks, ResultArgToks+NumToks); 472158115Sume 473158115Sume // In Microsoft-compatibility mode, we follow MSVC's preprocessing 474158115Sume // behavior by not considering single commas from nested macro 475158115Sume // expansions as argument separators. Set a flag on the token so we can 476158115Sume // test for this later when the macro expansion is processed. 477158115Sume if (PP.getLangOpts().MSVCCompat && NumToks == 1 && 478158115Sume ResultToks.back().is(tok::comma)) 479158115Sume ResultToks.back().setFlag(Token::IgnoredComma); 480158115Sume 481158115Sume // If the '##' came from expanding an argument, turn it into 'unknown' 482158115Sume // to avoid pasting. 483158115Sume for (Token &Tok : llvm::make_range(ResultToks.begin() + FirstResult, 484158115Sume ResultToks.end())) { 485158115Sume if (Tok.is(tok::hashhash)) 486158115Sume Tok.setKind(tok::unknown); 487158115Sume } 488158115Sume 489158115Sume if(ExpandLocStart.isValid()) { 490158115Sume updateLocForMacroArgTokens(CurTok.getLocation(), 491158115Sume ResultToks.begin()+FirstResult, 492158115Sume ResultToks.end()); 493158115Sume } 494158115Sume 495158115Sume // If any tokens were substituted from the argument, the whitespace 496158115Sume // before the first token should match the whitespace of the arg 497158115Sume // identifier. 498158115Sume ResultToks[FirstResult].setFlagValue(Token::LeadingSpace, 499158115Sume NextTokGetsSpace); 500158115Sume ResultToks[FirstResult].setFlagValue(Token::StartOfLine, false); 501158115Sume NextTokGetsSpace = false; 502158115Sume } else { 503158115Sume // We're creating a placeholder token. Usually this doesn't matter, 504158115Sume // but it can affect paste behavior when at the start or end of a 505158115Sume // __VA_OPT__. 506158115Sume if (NonEmptyPasteBefore) { 507158115Sume // We're imagining a placeholder token is inserted here. If this is 508158115Sume // the first token in a __VA_OPT__ after a ##, delete the ##. 509158115Sume assert(VCtx.isInVAOpt() && "should only happen inside a __VA_OPT__"); 510158115Sume VCtx.hasPlaceholderAfterHashhashAtStart(); 511158115Sume } 512158115Sume if (RParenAfter) 513158115Sume VCtx.hasPlaceholderBeforeRParen(); 514158115Sume } 515158115Sume continue; 516158115Sume } 517158115Sume 518158115Sume // Okay, we have a token that is either the LHS or RHS of a paste (##) 519158115Sume // argument. It gets substituted as its non-pre-expanded tokens. 520158115Sume const Token *ArgToks = ActualArgs->getUnexpArgument(ArgNo); 521158115Sume unsigned NumToks = MacroArgs::getArgLength(ArgToks); 522158115Sume if (NumToks) { // Not an empty argument? 523158115Sume bool VaArgsPseudoPaste = false; 524158115Sume // If this is the GNU ", ## __VA_ARGS__" extension, and we just learned 525158115Sume // that __VA_ARGS__ expands to multiple tokens, avoid a pasting error when 526158115Sume // the expander tries to paste ',' with the first token of the __VA_ARGS__ 527158115Sume // expansion. 528158115Sume if (NonEmptyPasteBefore && ResultToks.size() >= 2 && 529158115Sume ResultToks[ResultToks.size()-2].is(tok::comma) && 530158115Sume (unsigned)ArgNo == Macro->getNumParams()-1 && 531158115Sume Macro->isVariadic()) { 532158115Sume VaArgsPseudoPaste = true; 533158115Sume // Remove the paste operator, report use of the extension. 534158115Sume PP.Diag(ResultToks.pop_back_val().getLocation(), diag::ext_paste_comma); 535158115Sume } 536158115Sume 537158115Sume ResultToks.append(ArgToks, ArgToks+NumToks); 538 539 // If the '##' came from expanding an argument, turn it into 'unknown' 540 // to avoid pasting. 541 for (Token &Tok : llvm::make_range(ResultToks.end() - NumToks, 542 ResultToks.end())) { 543 if (Tok.is(tok::hashhash)) 544 Tok.setKind(tok::unknown); 545 } 546 547 if (ExpandLocStart.isValid()) { 548 updateLocForMacroArgTokens(CurTok.getLocation(), 549 ResultToks.end()-NumToks, ResultToks.end()); 550 } 551 552 // Transfer the leading whitespace information from the token 553 // (the macro argument) onto the first token of the 554 // expansion. Note that we don't do this for the GNU 555 // pseudo-paste extension ", ## __VA_ARGS__". 556 if (!VaArgsPseudoPaste) { 557 ResultToks[ResultToks.size() - NumToks].setFlagValue(Token::StartOfLine, 558 false); 559 ResultToks[ResultToks.size() - NumToks].setFlagValue( 560 Token::LeadingSpace, NextTokGetsSpace); 561 } 562 563 NextTokGetsSpace = false; 564 continue; 565 } 566 567 // If an empty argument is on the LHS or RHS of a paste, the standard (C99 568 // 6.10.3.3p2,3) calls for a bunch of placemarker stuff to occur. We 569 // implement this by eating ## operators when a LHS or RHS expands to 570 // empty. 571 if (PasteAfter) { 572 // Discard the argument token and skip (don't copy to the expansion 573 // buffer) the paste operator after it. 574 ++I; 575 continue; 576 } 577 578 if (RParenAfter) 579 VCtx.hasPlaceholderBeforeRParen(); 580 581 // If this is on the RHS of a paste operator, we've already copied the 582 // paste operator to the ResultToks list, unless the LHS was empty too. 583 // Remove it. 584 assert(PasteBefore); 585 if (NonEmptyPasteBefore) { 586 assert(ResultToks.back().is(tok::hashhash)); 587 // Do not remove the paste operator if it is the one before __VA_OPT__ 588 // (and we are still processing tokens within VA_OPT). We handle the case 589 // of removing the paste operator if __VA_OPT__ reduces to the notional 590 // placemarker above when we encounter the closing paren of VA_OPT. 591 if (!VCtx.isInVAOpt() || 592 ResultToks.size() > VCtx.getNumberOfTokensPriorToVAOpt()) 593 ResultToks.pop_back(); 594 else 595 VCtx.hasPlaceholderAfterHashhashAtStart(); 596 } 597 598 // If this is the __VA_ARGS__ token, and if the argument wasn't provided, 599 // and if the macro had at least one real argument, and if the token before 600 // the ## was a comma, remove the comma. This is a GCC extension which is 601 // disabled when using -std=c99. 602 if (ActualArgs->isVarargsElidedUse()) 603 MaybeRemoveCommaBeforeVaArgs(ResultToks, 604 /*HasPasteOperator=*/true, 605 Macro, ArgNo, PP); 606 } 607 608 // If anything changed, install this as the new Tokens list. 609 if (MadeChange) { 610 assert(!OwnsTokens && "This would leak if we already own the token list"); 611 // This is deleted in the dtor. 612 NumTokens = ResultToks.size(); 613 // The tokens will be added to Preprocessor's cache and will be removed 614 // when this TokenLexer finishes lexing them. 615 Tokens = PP.cacheMacroExpandedTokens(this, ResultToks); 616 617 // The preprocessor cache of macro expanded tokens owns these tokens,not us. 618 OwnsTokens = false; 619 } 620} 621 622/// Checks if two tokens form wide string literal. 623static bool isWideStringLiteralFromMacro(const Token &FirstTok, 624 const Token &SecondTok) { 625 return FirstTok.is(tok::identifier) && 626 FirstTok.getIdentifierInfo()->isStr("L") && SecondTok.isLiteral() && 627 SecondTok.stringifiedInMacro(); 628} 629 630/// Lex - Lex and return a token from this macro stream. 631bool TokenLexer::Lex(Token &Tok) { 632 // Lexing off the end of the macro, pop this macro off the expansion stack. 633 if (isAtEnd()) { 634 // If this is a macro (not a token stream), mark the macro enabled now 635 // that it is no longer being expanded. 636 if (Macro) Macro->EnableMacro(); 637 638 Tok.startToken(); 639 Tok.setFlagValue(Token::StartOfLine , AtStartOfLine); 640 Tok.setFlagValue(Token::LeadingSpace, HasLeadingSpace || NextTokGetsSpace); 641 if (CurTokenIdx == 0) 642 Tok.setFlag(Token::LeadingEmptyMacro); 643 return PP.HandleEndOfTokenLexer(Tok); 644 } 645 646 SourceManager &SM = PP.getSourceManager(); 647 648 // If this is the first token of the expanded result, we inherit spacing 649 // properties later. 650 bool isFirstToken = CurTokenIdx == 0; 651 652 // Get the next token to return. 653 Tok = Tokens[CurTokenIdx++]; 654 if (IsReinject) 655 Tok.setFlag(Token::IsReinjected); 656 657 bool TokenIsFromPaste = false; 658 659 // If this token is followed by a token paste (##) operator, paste the tokens! 660 // Note that ## is a normal token when not expanding a macro. 661 if (!isAtEnd() && Macro && 662 (Tokens[CurTokenIdx].is(tok::hashhash) || 663 // Special processing of L#x macros in -fms-compatibility mode. 664 // Microsoft compiler is able to form a wide string literal from 665 // 'L#macro_arg' construct in a function-like macro. 666 (PP.getLangOpts().MSVCCompat && 667 isWideStringLiteralFromMacro(Tok, Tokens[CurTokenIdx])))) { 668 // When handling the microsoft /##/ extension, the final token is 669 // returned by pasteTokens, not the pasted token. 670 if (pasteTokens(Tok)) 671 return true; 672 673 TokenIsFromPaste = true; 674 } 675 676 // The token's current location indicate where the token was lexed from. We 677 // need this information to compute the spelling of the token, but any 678 // diagnostics for the expanded token should appear as if they came from 679 // ExpansionLoc. Pull this information together into a new SourceLocation 680 // that captures all of this. 681 if (ExpandLocStart.isValid() && // Don't do this for token streams. 682 // Check that the token's location was not already set properly. 683 SM.isBeforeInSLocAddrSpace(Tok.getLocation(), MacroStartSLocOffset)) { 684 SourceLocation instLoc; 685 if (Tok.is(tok::comment)) { 686 instLoc = SM.createExpansionLoc(Tok.getLocation(), 687 ExpandLocStart, 688 ExpandLocEnd, 689 Tok.getLength()); 690 } else { 691 instLoc = getExpansionLocForMacroDefLoc(Tok.getLocation()); 692 } 693 694 Tok.setLocation(instLoc); 695 } 696 697 // If this is the first token, set the lexical properties of the token to 698 // match the lexical properties of the macro identifier. 699 if (isFirstToken) { 700 Tok.setFlagValue(Token::StartOfLine , AtStartOfLine); 701 Tok.setFlagValue(Token::LeadingSpace, HasLeadingSpace); 702 } else { 703 // If this is not the first token, we may still need to pass through 704 // leading whitespace if we've expanded a macro. 705 if (AtStartOfLine) Tok.setFlag(Token::StartOfLine); 706 if (HasLeadingSpace) Tok.setFlag(Token::LeadingSpace); 707 } 708 AtStartOfLine = false; 709 HasLeadingSpace = false; 710 711 // Handle recursive expansion! 712 if (!Tok.isAnnotation() && Tok.getIdentifierInfo() != nullptr) { 713 // Change the kind of this identifier to the appropriate token kind, e.g. 714 // turning "for" into a keyword. 715 IdentifierInfo *II = Tok.getIdentifierInfo(); 716 Tok.setKind(II->getTokenID()); 717 718 // If this identifier was poisoned and from a paste, emit an error. This 719 // won't be handled by Preprocessor::HandleIdentifier because this is coming 720 // from a macro expansion. 721 if (II->isPoisoned() && TokenIsFromPaste) { 722 PP.HandlePoisonedIdentifier(Tok); 723 } 724 725 if (!DisableMacroExpansion && II->isHandleIdentifierCase()) 726 return PP.HandleIdentifier(Tok); 727 } 728 729 // Otherwise, return a normal token. 730 return true; 731} 732 733bool TokenLexer::pasteTokens(Token &Tok) { 734 return pasteTokens(Tok, llvm::makeArrayRef(Tokens, NumTokens), CurTokenIdx); 735} 736 737/// LHSTok is the LHS of a ## operator, and CurTokenIdx is the ## 738/// operator. Read the ## and RHS, and paste the LHS/RHS together. If there 739/// are more ## after it, chomp them iteratively. Return the result as LHSTok. 740/// If this returns true, the caller should immediately return the token. 741bool TokenLexer::pasteTokens(Token &LHSTok, ArrayRef<Token> TokenStream, 742 unsigned int &CurIdx) { 743 assert(CurIdx > 0 && "## can not be the first token within tokens"); 744 assert((TokenStream[CurIdx].is(tok::hashhash) || 745 (PP.getLangOpts().MSVCCompat && 746 isWideStringLiteralFromMacro(LHSTok, TokenStream[CurIdx]))) && 747 "Token at this Index must be ## or part of the MSVC 'L " 748 "#macro-arg' pasting pair"); 749 750 // MSVC: If previous token was pasted, this must be a recovery from an invalid 751 // paste operation. Ignore spaces before this token to mimic MSVC output. 752 // Required for generating valid UUID strings in some MS headers. 753 if (PP.getLangOpts().MicrosoftExt && (CurIdx >= 2) && 754 TokenStream[CurIdx - 2].is(tok::hashhash)) 755 LHSTok.clearFlag(Token::LeadingSpace); 756 757 SmallString<128> Buffer; 758 const char *ResultTokStrPtr = nullptr; 759 SourceLocation StartLoc = LHSTok.getLocation(); 760 SourceLocation PasteOpLoc; 761 762 auto IsAtEnd = [&TokenStream, &CurIdx] { 763 return TokenStream.size() == CurIdx; 764 }; 765 766 do { 767 // Consume the ## operator if any. 768 PasteOpLoc = TokenStream[CurIdx].getLocation(); 769 if (TokenStream[CurIdx].is(tok::hashhash)) 770 ++CurIdx; 771 assert(!IsAtEnd() && "No token on the RHS of a paste operator!"); 772 773 // Get the RHS token. 774 const Token &RHS = TokenStream[CurIdx]; 775 776 // Allocate space for the result token. This is guaranteed to be enough for 777 // the two tokens. 778 Buffer.resize(LHSTok.getLength() + RHS.getLength()); 779 780 // Get the spelling of the LHS token in Buffer. 781 const char *BufPtr = &Buffer[0]; 782 bool Invalid = false; 783 unsigned LHSLen = PP.getSpelling(LHSTok, BufPtr, &Invalid); 784 if (BufPtr != &Buffer[0]) // Really, we want the chars in Buffer! 785 memcpy(&Buffer[0], BufPtr, LHSLen); 786 if (Invalid) 787 return true; 788 789 BufPtr = Buffer.data() + LHSLen; 790 unsigned RHSLen = PP.getSpelling(RHS, BufPtr, &Invalid); 791 if (Invalid) 792 return true; 793 if (RHSLen && BufPtr != &Buffer[LHSLen]) 794 // Really, we want the chars in Buffer! 795 memcpy(&Buffer[LHSLen], BufPtr, RHSLen); 796 797 // Trim excess space. 798 Buffer.resize(LHSLen+RHSLen); 799 800 // Plop the pasted result (including the trailing newline and null) into a 801 // scratch buffer where we can lex it. 802 Token ResultTokTmp; 803 ResultTokTmp.startToken(); 804 805 // Claim that the tmp token is a string_literal so that we can get the 806 // character pointer back from CreateString in getLiteralData(). 807 ResultTokTmp.setKind(tok::string_literal); 808 PP.CreateString(Buffer, ResultTokTmp); 809 SourceLocation ResultTokLoc = ResultTokTmp.getLocation(); 810 ResultTokStrPtr = ResultTokTmp.getLiteralData(); 811 812 // Lex the resultant pasted token into Result. 813 Token Result; 814 815 if (LHSTok.isAnyIdentifier() && RHS.isAnyIdentifier()) { 816 // Common paste case: identifier+identifier = identifier. Avoid creating 817 // a lexer and other overhead. 818 PP.IncrementPasteCounter(true); 819 Result.startToken(); 820 Result.setKind(tok::raw_identifier); 821 Result.setRawIdentifierData(ResultTokStrPtr); 822 Result.setLocation(ResultTokLoc); 823 Result.setLength(LHSLen+RHSLen); 824 } else { 825 PP.IncrementPasteCounter(false); 826 827 assert(ResultTokLoc.isFileID() && 828 "Should be a raw location into scratch buffer"); 829 SourceManager &SourceMgr = PP.getSourceManager(); 830 FileID LocFileID = SourceMgr.getFileID(ResultTokLoc); 831 832 bool Invalid = false; 833 const char *ScratchBufStart 834 = SourceMgr.getBufferData(LocFileID, &Invalid).data(); 835 if (Invalid) 836 return false; 837 838 // Make a lexer to lex this string from. Lex just this one token. 839 // Make a lexer object so that we lex and expand the paste result. 840 Lexer TL(SourceMgr.getLocForStartOfFile(LocFileID), 841 PP.getLangOpts(), ScratchBufStart, 842 ResultTokStrPtr, ResultTokStrPtr+LHSLen+RHSLen); 843 844 // Lex a token in raw mode. This way it won't look up identifiers 845 // automatically, lexing off the end will return an eof token, and 846 // warnings are disabled. This returns true if the result token is the 847 // entire buffer. 848 bool isInvalid = !TL.LexFromRawLexer(Result); 849 850 // If we got an EOF token, we didn't form even ONE token. For example, we 851 // did "/ ## /" to get "//". 852 isInvalid |= Result.is(tok::eof); 853 854 // If pasting the two tokens didn't form a full new token, this is an 855 // error. This occurs with "x ## +" and other stuff. Return with LHSTok 856 // unmodified and with RHS as the next token to lex. 857 if (isInvalid) { 858 // Explicitly convert the token location to have proper expansion 859 // information so that the user knows where it came from. 860 SourceManager &SM = PP.getSourceManager(); 861 SourceLocation Loc = 862 SM.createExpansionLoc(PasteOpLoc, ExpandLocStart, ExpandLocEnd, 2); 863 864 // Test for the Microsoft extension of /##/ turning into // here on the 865 // error path. 866 if (PP.getLangOpts().MicrosoftExt && LHSTok.is(tok::slash) && 867 RHS.is(tok::slash)) { 868 HandleMicrosoftCommentPaste(LHSTok, Loc); 869 return true; 870 } 871 872 // Do not emit the error when preprocessing assembler code. 873 if (!PP.getLangOpts().AsmPreprocessor) { 874 // If we're in microsoft extensions mode, downgrade this from a hard 875 // error to an extension that defaults to an error. This allows 876 // disabling it. 877 PP.Diag(Loc, PP.getLangOpts().MicrosoftExt ? diag::ext_pp_bad_paste_ms 878 : diag::err_pp_bad_paste) 879 << Buffer; 880 } 881 882 // An error has occurred so exit loop. 883 break; 884 } 885 886 // Turn ## into 'unknown' to avoid # ## # from looking like a paste 887 // operator. 888 if (Result.is(tok::hashhash)) 889 Result.setKind(tok::unknown); 890 } 891 892 // Transfer properties of the LHS over the Result. 893 Result.setFlagValue(Token::StartOfLine , LHSTok.isAtStartOfLine()); 894 Result.setFlagValue(Token::LeadingSpace, LHSTok.hasLeadingSpace()); 895 896 // Finally, replace LHS with the result, consume the RHS, and iterate. 897 ++CurIdx; 898 LHSTok = Result; 899 } while (!IsAtEnd() && TokenStream[CurIdx].is(tok::hashhash)); 900 901 SourceLocation EndLoc = TokenStream[CurIdx - 1].getLocation(); 902 903 // The token's current location indicate where the token was lexed from. We 904 // need this information to compute the spelling of the token, but any 905 // diagnostics for the expanded token should appear as if the token was 906 // expanded from the full ## expression. Pull this information together into 907 // a new SourceLocation that captures all of this. 908 SourceManager &SM = PP.getSourceManager(); 909 if (StartLoc.isFileID()) 910 StartLoc = getExpansionLocForMacroDefLoc(StartLoc); 911 if (EndLoc.isFileID()) 912 EndLoc = getExpansionLocForMacroDefLoc(EndLoc); 913 FileID MacroFID = SM.getFileID(MacroExpansionStart); 914 while (SM.getFileID(StartLoc) != MacroFID) 915 StartLoc = SM.getImmediateExpansionRange(StartLoc).getBegin(); 916 while (SM.getFileID(EndLoc) != MacroFID) 917 EndLoc = SM.getImmediateExpansionRange(EndLoc).getEnd(); 918 919 LHSTok.setLocation(SM.createExpansionLoc(LHSTok.getLocation(), StartLoc, EndLoc, 920 LHSTok.getLength())); 921 922 // Now that we got the result token, it will be subject to expansion. Since 923 // token pasting re-lexes the result token in raw mode, identifier information 924 // isn't looked up. As such, if the result is an identifier, look up id info. 925 if (LHSTok.is(tok::raw_identifier)) { 926 // Look up the identifier info for the token. We disabled identifier lookup 927 // by saying we're skipping contents, so we need to do this manually. 928 PP.LookUpIdentifierInfo(LHSTok); 929 } 930 return false; 931} 932 933/// isNextTokenLParen - If the next token lexed will pop this macro off the 934/// expansion stack, return 2. If the next unexpanded token is a '(', return 935/// 1, otherwise return 0. 936unsigned TokenLexer::isNextTokenLParen() const { 937 // Out of tokens? 938 if (isAtEnd()) 939 return 2; 940 return Tokens[CurTokenIdx].is(tok::l_paren); 941} 942 943/// isParsingPreprocessorDirective - Return true if we are in the middle of a 944/// preprocessor directive. 945bool TokenLexer::isParsingPreprocessorDirective() const { 946 return Tokens[NumTokens-1].is(tok::eod) && !isAtEnd(); 947} 948 949/// HandleMicrosoftCommentPaste - In microsoft compatibility mode, /##/ pastes 950/// together to form a comment that comments out everything in the current 951/// macro, other active macros, and anything left on the current physical 952/// source line of the expanded buffer. Handle this by returning the 953/// first token on the next line. 954void TokenLexer::HandleMicrosoftCommentPaste(Token &Tok, SourceLocation OpLoc) { 955 PP.Diag(OpLoc, diag::ext_comment_paste_microsoft); 956 957 // We 'comment out' the rest of this macro by just ignoring the rest of the 958 // tokens that have not been lexed yet, if any. 959 960 // Since this must be a macro, mark the macro enabled now that it is no longer 961 // being expanded. 962 assert(Macro && "Token streams can't paste comments"); 963 Macro->EnableMacro(); 964 965 PP.HandleMicrosoftCommentPaste(Tok); 966} 967 968/// If \arg loc is a file ID and points inside the current macro 969/// definition, returns the appropriate source location pointing at the 970/// macro expansion source location entry, otherwise it returns an invalid 971/// SourceLocation. 972SourceLocation 973TokenLexer::getExpansionLocForMacroDefLoc(SourceLocation loc) const { 974 assert(ExpandLocStart.isValid() && MacroExpansionStart.isValid() && 975 "Not appropriate for token streams"); 976 assert(loc.isValid() && loc.isFileID()); 977 978 SourceManager &SM = PP.getSourceManager(); 979 assert(SM.isInSLocAddrSpace(loc, MacroDefStart, MacroDefLength) && 980 "Expected loc to come from the macro definition"); 981 982 unsigned relativeOffset = 0; 983 SM.isInSLocAddrSpace(loc, MacroDefStart, MacroDefLength, &relativeOffset); 984 return MacroExpansionStart.getLocWithOffset(relativeOffset); 985} 986 987/// Finds the tokens that are consecutive (from the same FileID) 988/// creates a single SLocEntry, and assigns SourceLocations to each token that 989/// point to that SLocEntry. e.g for 990/// assert(foo == bar); 991/// There will be a single SLocEntry for the "foo == bar" chunk and locations 992/// for the 'foo', '==', 'bar' tokens will point inside that chunk. 993/// 994/// \arg begin_tokens will be updated to a position past all the found 995/// consecutive tokens. 996static void updateConsecutiveMacroArgTokens(SourceManager &SM, 997 SourceLocation InstLoc, 998 Token *&begin_tokens, 999 Token * end_tokens) { 1000 assert(begin_tokens < end_tokens); 1001 1002 SourceLocation FirstLoc = begin_tokens->getLocation(); 1003 SourceLocation CurLoc = FirstLoc; 1004 1005 // Compare the source location offset of tokens and group together tokens that 1006 // are close, even if their locations point to different FileIDs. e.g. 1007 // 1008 // |bar | foo | cake | (3 tokens from 3 consecutive FileIDs) 1009 // ^ ^ 1010 // |bar foo cake| (one SLocEntry chunk for all tokens) 1011 // 1012 // we can perform this "merge" since the token's spelling location depends 1013 // on the relative offset. 1014 1015 Token *NextTok = begin_tokens + 1; 1016 for (; NextTok < end_tokens; ++NextTok) { 1017 SourceLocation NextLoc = NextTok->getLocation(); 1018 if (CurLoc.isFileID() != NextLoc.isFileID()) 1019 break; // Token from different kind of FileID. 1020 1021 int RelOffs; 1022 if (!SM.isInSameSLocAddrSpace(CurLoc, NextLoc, &RelOffs)) 1023 break; // Token from different local/loaded location. 1024 // Check that token is not before the previous token or more than 50 1025 // "characters" away. 1026 if (RelOffs < 0 || RelOffs > 50) 1027 break; 1028 1029 if (CurLoc.isMacroID() && !SM.isWrittenInSameFile(CurLoc, NextLoc)) 1030 break; // Token from a different macro. 1031 1032 CurLoc = NextLoc; 1033 } 1034 1035 // For the consecutive tokens, find the length of the SLocEntry to contain 1036 // all of them. 1037 Token &LastConsecutiveTok = *(NextTok-1); 1038 int LastRelOffs = 0; 1039 SM.isInSameSLocAddrSpace(FirstLoc, LastConsecutiveTok.getLocation(), 1040 &LastRelOffs); 1041 unsigned FullLength = LastRelOffs + LastConsecutiveTok.getLength(); 1042 1043 // Create a macro expansion SLocEntry that will "contain" all of the tokens. 1044 SourceLocation Expansion = 1045 SM.createMacroArgExpansionLoc(FirstLoc, InstLoc,FullLength); 1046 1047 // Change the location of the tokens from the spelling location to the new 1048 // expanded location. 1049 for (; begin_tokens < NextTok; ++begin_tokens) { 1050 Token &Tok = *begin_tokens; 1051 int RelOffs = 0; 1052 SM.isInSameSLocAddrSpace(FirstLoc, Tok.getLocation(), &RelOffs); 1053 Tok.setLocation(Expansion.getLocWithOffset(RelOffs)); 1054 } 1055} 1056 1057/// Creates SLocEntries and updates the locations of macro argument 1058/// tokens to their new expanded locations. 1059/// 1060/// \param ArgIdSpellLoc the location of the macro argument id inside the macro 1061/// definition. 1062void TokenLexer::updateLocForMacroArgTokens(SourceLocation ArgIdSpellLoc, 1063 Token *begin_tokens, 1064 Token *end_tokens) { 1065 SourceManager &SM = PP.getSourceManager(); 1066 1067 SourceLocation InstLoc = 1068 getExpansionLocForMacroDefLoc(ArgIdSpellLoc); 1069 1070 while (begin_tokens < end_tokens) { 1071 // If there's only one token just create a SLocEntry for it. 1072 if (end_tokens - begin_tokens == 1) { 1073 Token &Tok = *begin_tokens; 1074 Tok.setLocation(SM.createMacroArgExpansionLoc(Tok.getLocation(), 1075 InstLoc, 1076 Tok.getLength())); 1077 return; 1078 } 1079 1080 updateConsecutiveMacroArgTokens(SM, InstLoc, begin_tokens, end_tokens); 1081 } 1082} 1083 1084void TokenLexer::PropagateLineStartLeadingSpaceInfo(Token &Result) { 1085 AtStartOfLine = Result.isAtStartOfLine(); 1086 HasLeadingSpace = Result.hasLeadingSpace(); 1087} 1088