1 //===- TokenLexer.cpp - Lex from a token stream ---------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements the TokenLexer interface.
12 //===----------------------------------------------------------------------===//
14 #include "clang/Lex/TokenLexer.h"
15 #include "clang/Basic/Diagnostic.h"
16 #include "clang/Basic/IdentifierTable.h"
17 #include "clang/Basic/LangOptions.h"
18 #include "clang/Basic/SourceLocation.h"
19 #include "clang/Basic/SourceManager.h"
20 #include "clang/Basic/TokenKinds.h"
21 #include "clang/Lex/LexDiagnostic.h"
22 #include "clang/Lex/Lexer.h"
23 #include "clang/Lex/MacroArgs.h"
24 #include "clang/Lex/MacroInfo.h"
25 #include "clang/Lex/Preprocessor.h"
26 #include "clang/Lex/Token.h"
27 #include "clang/Lex/VariadicMacroSupport.h"
28 #include "llvm/ADT/ArrayRef.h"
29 #include "llvm/ADT/SmallString.h"
30 #include "llvm/ADT/SmallVector.h"
31 #include "llvm/ADT/iterator_range.h"
35 using namespace clang;
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,
41 // If the client is reusing a TokenLexer, make sure to free any memory
42 // associated with it.
49 ExpandLocStart = Tok.getLocation();
51 AtStartOfLine = Tok.isAtStartOfLine();
52 HasLeadingSpace = Tok.hasLeadingSpace();
53 NextTokGetsSpace = false;
54 Tokens = &*Macro->tokens_begin();
56 DisableMacroExpansion = false;
57 NumTokens = Macro->tokens_end()-Macro->tokens_begin();
58 MacroExpansionStart = SourceLocation();
60 SourceManager &SM = PP.getSourceManager();
61 MacroStartSLocOffset = SM.getNextLocalOffset();
64 assert(Tokens[0].getLocation().isValid());
65 assert((Tokens[0].getLocation().isFileID() || Tokens[0].is(tok::comment)) &&
66 "Macro defined in macro?");
67 assert(ExpandLocStart.isValid());
69 // Reserve a source location entry chunk for the length of the macro
70 // definition. Tokens that get lexed directly from the definition will
71 // have their locations pointing inside this chunk. This is to avoid
72 // creating separate source location entries for each token.
73 MacroDefStart = SM.getExpansionLoc(Tokens[0].getLocation());
74 MacroDefLength = Macro->getDefinitionLength(SM);
75 MacroExpansionStart = SM.createExpansionLoc(MacroDefStart,
81 // If this is a function-like macro, expand the arguments and change
82 // Tokens to point to the expanded tokens.
83 if (Macro->isFunctionLike() && Macro->getNumParams())
84 ExpandFunctionArguments();
86 // Mark the macro as currently disabled, so that it is not recursively
87 // expanded. The macro must be disabled only after argument pre-expansion of
88 // function-like macro arguments occurs.
89 Macro->DisableMacro();
92 /// Create a TokenLexer for the specified token stream. This does not
93 /// take ownership of the specified token vector.
94 void TokenLexer::Init(const Token *TokArray, unsigned NumToks,
95 bool disableMacroExpansion, bool ownsTokens) {
96 // If the client is reusing a TokenLexer, make sure to free any memory
97 // associated with it.
101 ActualArgs = nullptr;
103 OwnsTokens = ownsTokens;
104 DisableMacroExpansion = disableMacroExpansion;
107 ExpandLocStart = ExpandLocEnd = SourceLocation();
108 AtStartOfLine = false;
109 HasLeadingSpace = false;
110 NextTokGetsSpace = false;
111 MacroExpansionStart = SourceLocation();
113 // Set HasLeadingSpace/AtStartOfLine so that the first token will be
114 // returned unmodified.
116 AtStartOfLine = TokArray[0].isAtStartOfLine();
117 HasLeadingSpace = TokArray[0].hasLeadingSpace();
121 void TokenLexer::destroy() {
122 // If this was a function-like macro that actually uses its arguments, delete
123 // the expanded tokens.
130 // TokenLexer owns its formal arguments.
131 if (ActualArgs) ActualArgs->destroy(PP);
134 bool TokenLexer::MaybeRemoveCommaBeforeVaArgs(
135 SmallVectorImpl<Token> &ResultToks, bool HasPasteOperator, MacroInfo *Macro,
136 unsigned MacroArgNo, Preprocessor &PP) {
137 // Is the macro argument __VA_ARGS__?
138 if (!Macro->isVariadic() || MacroArgNo != Macro->getNumParams()-1)
141 // In Microsoft-compatibility mode, a comma is removed in the expansion
142 // of " ... , __VA_ARGS__ " if __VA_ARGS__ is empty. This extension is
143 // not supported by gcc.
144 if (!HasPasteOperator && !PP.getLangOpts().MSVCCompat)
147 // GCC removes the comma in the expansion of " ... , ## __VA_ARGS__ " if
148 // __VA_ARGS__ is empty, but not in strict C99 mode where there are no
149 // named arguments, where it remains. In all other modes, including C99
150 // with GNU extensions, it is removed regardless of named arguments.
151 // Microsoft also appears to support this extension, unofficially.
152 if (PP.getLangOpts().C99 && !PP.getLangOpts().GNUMode
153 && Macro->getNumParams() < 2)
156 // Is a comma available to be removed?
157 if (ResultToks.empty() || !ResultToks.back().is(tok::comma))
160 // Issue an extension diagnostic for the paste operator.
161 if (HasPasteOperator)
162 PP.Diag(ResultToks.back().getLocation(), diag::ext_paste_comma);
165 ResultToks.pop_back();
167 if (!ResultToks.empty()) {
168 // If the comma was right after another paste (e.g. "X##,##__VA_ARGS__"),
169 // then removal of the comma should produce a placemarker token (in C99
170 // terms) which we model by popping off the previous ##, giving us a plain
171 // "X" when __VA_ARGS__ is empty.
172 if (ResultToks.back().is(tok::hashhash))
173 ResultToks.pop_back();
175 // Remember that this comma was elided.
176 ResultToks.back().setFlag(Token::CommaAfterElided);
179 // Never add a space, even if the comma, ##, or arg had a space.
180 NextTokGetsSpace = false;
184 void TokenLexer::stringifyVAOPTContents(
185 SmallVectorImpl<Token> &ResultToks, const VAOptExpansionContext &VCtx,
186 const SourceLocation VAOPTClosingParenLoc) {
187 const int NumToksPriorToVAOpt = VCtx.getNumberOfTokensPriorToVAOpt();
188 const unsigned int NumVAOptTokens = ResultToks.size() - NumToksPriorToVAOpt;
189 Token *const VAOPTTokens =
190 NumVAOptTokens ? &ResultToks[NumToksPriorToVAOpt] : nullptr;
192 SmallVector<Token, 64> ConcatenatedVAOPTResultToks;
193 // FIXME: Should we keep track within VCtx that we did or didnot
194 // encounter pasting - and only then perform this loop.
196 // Perform token pasting (concatenation) prior to stringization.
197 for (unsigned int CurTokenIdx = 0; CurTokenIdx != NumVAOptTokens;
199 if (VAOPTTokens[CurTokenIdx].is(tok::hashhash)) {
200 assert(CurTokenIdx != 0 &&
201 "Can not have __VAOPT__ contents begin with a ##");
202 Token &LHS = VAOPTTokens[CurTokenIdx - 1];
203 pasteTokens(LHS, llvm::makeArrayRef(VAOPTTokens, NumVAOptTokens),
205 // Replace the token prior to the first ## in this iteration.
206 ConcatenatedVAOPTResultToks.back() = LHS;
207 if (CurTokenIdx == NumVAOptTokens)
210 ConcatenatedVAOPTResultToks.push_back(VAOPTTokens[CurTokenIdx]);
213 ConcatenatedVAOPTResultToks.push_back(VCtx.getEOFTok());
214 // Get the SourceLocation that represents the start location within
215 // the macro definition that marks where this string is substituted
216 // into: i.e. the __VA_OPT__ and the ')' within the spelling of the
217 // macro definition, and use it to indicate that the stringified token
218 // was generated from that location.
219 const SourceLocation ExpansionLocStartWithinMacro =
220 getExpansionLocForMacroDefLoc(VCtx.getVAOptLoc());
221 const SourceLocation ExpansionLocEndWithinMacro =
222 getExpansionLocForMacroDefLoc(VAOPTClosingParenLoc);
224 Token StringifiedVAOPT = MacroArgs::StringifyArgument(
225 &ConcatenatedVAOPTResultToks[0], PP, VCtx.hasCharifyBefore() /*Charify*/,
226 ExpansionLocStartWithinMacro, ExpansionLocEndWithinMacro);
228 if (VCtx.getLeadingSpaceForStringifiedToken())
229 StringifiedVAOPT.setFlag(Token::LeadingSpace);
231 StringifiedVAOPT.setFlag(Token::StringifiedInMacro);
232 // Resize (shrink) the token stream to just capture this stringified token.
233 ResultToks.resize(NumToksPriorToVAOpt + 1);
234 ResultToks.back() = StringifiedVAOPT;
237 /// Expand the arguments of a function-like macro so that we can quickly
238 /// return preexpanded tokens from Tokens.
239 void TokenLexer::ExpandFunctionArguments() {
240 SmallVector<Token, 128> ResultToks;
242 // Loop through 'Tokens', expanding them into ResultToks. Keep
243 // track of whether we change anything. If not, no need to keep them. If so,
244 // we install the newly expanded sequence as the new 'Tokens' list.
245 bool MadeChange = false;
247 const bool CalledWithVariadicArguments =
248 ActualArgs->invokedWithVariadicArgument(Macro);
250 VAOptExpansionContext VCtx(PP);
252 for (unsigned I = 0, E = NumTokens; I != E; ++I) {
253 const Token &CurTok = Tokens[I];
254 // We don't want a space for the next token after a paste
255 // operator. In valid code, the token will get smooshed onto the
256 // preceding one anyway. In assembler-with-cpp mode, invalid
257 // pastes are allowed through: in this case, we do not want the
258 // extra whitespace to be added. For example, we want ". ## foo"
259 // -> ".foo" not ". foo".
260 if (I != 0 && !Tokens[I-1].is(tok::hashhash) && CurTok.hasLeadingSpace())
261 NextTokGetsSpace = true;
263 if (VCtx.isVAOptToken(CurTok)) {
265 assert(Tokens[I + 1].is(tok::l_paren) &&
266 "__VA_OPT__ must be followed by '('");
268 ++I; // Skip the l_paren
269 VCtx.sawVAOptFollowedByOpeningParens(CurTok.getLocation(),
275 // We have entered into the __VA_OPT__ context, so handle tokens
277 if (VCtx.isInVAOpt()) {
278 // If we are about to process a token that is either an argument to
279 // __VA_OPT__ or its closing rparen, then:
280 // 1) If the token is the closing rparen that exits us out of __VA_OPT__,
281 // perform any necessary stringification or placemarker processing,
282 // and/or skip to the next token.
283 // 2) else if macro was invoked without variadic arguments skip this
285 // 3) else (macro was invoked with variadic arguments) process the token
288 if (Tokens[I].is(tok::l_paren))
289 VCtx.sawOpeningParen(Tokens[I].getLocation());
290 // Continue skipping tokens within __VA_OPT__ if the macro was not
291 // called with variadic arguments, else let the rest of the loop handle
292 // this token. Note sawClosingParen() returns true only if the r_paren matches
293 // the closing r_paren of the __VA_OPT__.
294 if (!Tokens[I].is(tok::r_paren) || !VCtx.sawClosingParen()) {
295 if (!CalledWithVariadicArguments) {
299 // ... else the macro was called with variadic arguments, and we do not
300 // have a closing rparen - so process this token normally.
302 // Current token is the closing r_paren which marks the end of the
303 // __VA_OPT__ invocation, so handle any place-marker pasting (if
304 // empty) by removing hashhash either before (if exists) or after. And
305 // also stringify the entire contents if VAOPT was preceded by a hash,
306 // but do so only after any token concatenation that needs to occur
307 // within the contents of VAOPT.
309 if (VCtx.hasStringifyOrCharifyBefore()) {
310 // Replace all the tokens just added from within VAOPT into a single
311 // stringified token. This requires token-pasting to eagerly occur
312 // within these tokens. If either the contents of VAOPT were empty
313 // or the macro wasn't called with any variadic arguments, the result
314 // is a token that represents an empty string.
315 stringifyVAOPTContents(ResultToks, VCtx,
316 /*ClosingParenLoc*/ Tokens[I].getLocation());
318 } else if (/*No tokens within VAOPT*/ !(
319 ResultToks.size() - VCtx.getNumberOfTokensPriorToVAOpt())) {
320 // Treat VAOPT as a placemarker token. Eat either the '##' before the
321 // RHS/VAOPT (if one exists, suggesting that the LHS (if any) to that
322 // hashhash was not a placemarker) or the '##'
323 // after VAOPT, but not both.
325 if (ResultToks.size() && ResultToks.back().is(tok::hashhash)) {
326 ResultToks.pop_back();
327 } else if ((I + 1 != E) && Tokens[I + 1].is(tok::hashhash)) {
328 ++I; // Skip the following hashhash.
332 // We processed __VA_OPT__'s closing paren (and the exit out of
333 // __VA_OPT__), so skip to the next token.
338 // If we found the stringify operator, get the argument stringified. The
339 // preprocessor already verified that the following token is a macro
340 // parameter or __VA_OPT__ when the #define was lexed.
342 if (CurTok.isOneOf(tok::hash, tok::hashat)) {
343 int ArgNo = Macro->getParameterNum(Tokens[I+1].getIdentifierInfo());
344 assert((ArgNo != -1 || VCtx.isVAOptToken(Tokens[I + 1])) &&
345 "Token following # is not an argument or __VA_OPT__!");
348 // Handle the __VA_OPT__ case.
349 VCtx.sawHashOrHashAtBefore(NextTokGetsSpace,
350 CurTok.is(tok::hashat));
353 // Else handle the simple argument case.
354 SourceLocation ExpansionLocStart =
355 getExpansionLocForMacroDefLoc(CurTok.getLocation());
356 SourceLocation ExpansionLocEnd =
357 getExpansionLocForMacroDefLoc(Tokens[I+1].getLocation());
360 if (CurTok.is(tok::hash)) // Stringify
361 Res = ActualArgs->getStringifiedArgument(ArgNo, PP,
365 // 'charify': don't bother caching these.
366 Res = MacroArgs::StringifyArgument(ActualArgs->getUnexpArgument(ArgNo),
371 Res.setFlag(Token::StringifiedInMacro);
373 // The stringified/charified string leading space flag gets set to match
374 // the #/#@ operator.
375 if (NextTokGetsSpace)
376 Res.setFlag(Token::LeadingSpace);
378 ResultToks.push_back(Res);
380 ++I; // Skip arg name.
381 NextTokGetsSpace = false;
385 // Find out if there is a paste (##) operator before or after the token.
386 bool NonEmptyPasteBefore =
387 !ResultToks.empty() && ResultToks.back().is(tok::hashhash);
388 bool PasteBefore = I != 0 && Tokens[I-1].is(tok::hashhash);
389 bool PasteAfter = I+1 != E && Tokens[I+1].is(tok::hashhash);
391 assert((!NonEmptyPasteBefore || PasteBefore || VCtx.isInVAOpt()) &&
392 "unexpected ## in ResultToks");
394 // Otherwise, if this is not an argument token, just add the token to the
396 IdentifierInfo *II = CurTok.getIdentifierInfo();
397 int ArgNo = II ? Macro->getParameterNum(II) : -1;
399 // This isn't an argument, just add it.
400 ResultToks.push_back(CurTok);
402 if (NextTokGetsSpace) {
403 ResultToks.back().setFlag(Token::LeadingSpace);
404 NextTokGetsSpace = false;
405 } else if (PasteBefore && !NonEmptyPasteBefore)
406 ResultToks.back().clearFlag(Token::LeadingSpace);
411 // An argument is expanded somehow, the result is different than the
415 // Otherwise, this is a use of the argument.
417 // In Microsoft mode, remove the comma before __VA_ARGS__ to ensure there
418 // are no trailing commas if __VA_ARGS__ is empty.
419 if (!PasteBefore && ActualArgs->isVarargsElidedUse() &&
420 MaybeRemoveCommaBeforeVaArgs(ResultToks,
421 /*HasPasteOperator=*/false,
425 // If it is not the LHS/RHS of a ## operator, we must pre-expand the
426 // argument and substitute the expanded tokens into the result. This is
428 if (!PasteBefore && !PasteAfter) {
429 const Token *ResultArgToks;
431 // Only preexpand the argument if it could possibly need it. This
432 // avoids some work in common cases.
433 const Token *ArgTok = ActualArgs->getUnexpArgument(ArgNo);
434 if (ActualArgs->ArgNeedsPreexpansion(ArgTok, PP))
435 ResultArgToks = &ActualArgs->getPreExpArgument(ArgNo, PP)[0];
437 ResultArgToks = ArgTok; // Use non-preexpanded tokens.
439 // If the arg token expanded into anything, append it.
440 if (ResultArgToks->isNot(tok::eof)) {
441 size_t FirstResult = ResultToks.size();
442 unsigned NumToks = MacroArgs::getArgLength(ResultArgToks);
443 ResultToks.append(ResultArgToks, ResultArgToks+NumToks);
445 // In Microsoft-compatibility mode, we follow MSVC's preprocessing
446 // behavior by not considering single commas from nested macro
447 // expansions as argument separators. Set a flag on the token so we can
448 // test for this later when the macro expansion is processed.
449 if (PP.getLangOpts().MSVCCompat && NumToks == 1 &&
450 ResultToks.back().is(tok::comma))
451 ResultToks.back().setFlag(Token::IgnoredComma);
453 // If the '##' came from expanding an argument, turn it into 'unknown'
455 for (Token &Tok : llvm::make_range(ResultToks.begin() + FirstResult,
457 if (Tok.is(tok::hashhash))
458 Tok.setKind(tok::unknown);
461 if(ExpandLocStart.isValid()) {
462 updateLocForMacroArgTokens(CurTok.getLocation(),
463 ResultToks.begin()+FirstResult,
467 // If any tokens were substituted from the argument, the whitespace
468 // before the first token should match the whitespace of the arg
470 ResultToks[FirstResult].setFlagValue(Token::LeadingSpace,
472 ResultToks[FirstResult].setFlagValue(Token::StartOfLine, false);
473 NextTokGetsSpace = false;
478 // Okay, we have a token that is either the LHS or RHS of a paste (##)
479 // argument. It gets substituted as its non-pre-expanded tokens.
480 const Token *ArgToks = ActualArgs->getUnexpArgument(ArgNo);
481 unsigned NumToks = MacroArgs::getArgLength(ArgToks);
482 if (NumToks) { // Not an empty argument?
483 bool VaArgsPseudoPaste = false;
484 // If this is the GNU ", ## __VA_ARGS__" extension, and we just learned
485 // that __VA_ARGS__ expands to multiple tokens, avoid a pasting error when
486 // the expander trys to paste ',' with the first token of the __VA_ARGS__
488 if (NonEmptyPasteBefore && ResultToks.size() >= 2 &&
489 ResultToks[ResultToks.size()-2].is(tok::comma) &&
490 (unsigned)ArgNo == Macro->getNumParams()-1 &&
491 Macro->isVariadic()) {
492 VaArgsPseudoPaste = true;
493 // Remove the paste operator, report use of the extension.
494 PP.Diag(ResultToks.pop_back_val().getLocation(), diag::ext_paste_comma);
497 ResultToks.append(ArgToks, ArgToks+NumToks);
499 // If the '##' came from expanding an argument, turn it into 'unknown'
501 for (Token &Tok : llvm::make_range(ResultToks.end() - NumToks,
503 if (Tok.is(tok::hashhash))
504 Tok.setKind(tok::unknown);
507 if (ExpandLocStart.isValid()) {
508 updateLocForMacroArgTokens(CurTok.getLocation(),
509 ResultToks.end()-NumToks, ResultToks.end());
512 // Transfer the leading whitespace information from the token
513 // (the macro argument) onto the first token of the
514 // expansion. Note that we don't do this for the GNU
515 // pseudo-paste extension ", ## __VA_ARGS__".
516 if (!VaArgsPseudoPaste) {
517 ResultToks[ResultToks.size() - NumToks].setFlagValue(Token::StartOfLine,
519 ResultToks[ResultToks.size() - NumToks].setFlagValue(
520 Token::LeadingSpace, NextTokGetsSpace);
523 NextTokGetsSpace = false;
527 // If an empty argument is on the LHS or RHS of a paste, the standard (C99
528 // 6.10.3.3p2,3) calls for a bunch of placemarker stuff to occur. We
529 // implement this by eating ## operators when a LHS or RHS expands to
532 // Discard the argument token and skip (don't copy to the expansion
533 // buffer) the paste operator after it.
538 // If this is on the RHS of a paste operator, we've already copied the
539 // paste operator to the ResultToks list, unless the LHS was empty too.
542 if (NonEmptyPasteBefore) {
543 assert(ResultToks.back().is(tok::hashhash));
544 // Do not remove the paste operator if it is the one before __VA_OPT__
545 // (and we are still processing tokens within VA_OPT). We handle the case
546 // of removing the paste operator if __VA_OPT__ reduces to the notional
547 // placemarker above when we encounter the closing paren of VA_OPT.
548 if (!VCtx.isInVAOpt() ||
549 ResultToks.size() > VCtx.getNumberOfTokensPriorToVAOpt())
550 ResultToks.pop_back();
553 // If this is the __VA_ARGS__ token, and if the argument wasn't provided,
554 // and if the macro had at least one real argument, and if the token before
555 // the ## was a comma, remove the comma. This is a GCC extension which is
556 // disabled when using -std=c99.
557 if (ActualArgs->isVarargsElidedUse())
558 MaybeRemoveCommaBeforeVaArgs(ResultToks,
559 /*HasPasteOperator=*/true,
563 // If anything changed, install this as the new Tokens list.
565 assert(!OwnsTokens && "This would leak if we already own the token list");
566 // This is deleted in the dtor.
567 NumTokens = ResultToks.size();
568 // The tokens will be added to Preprocessor's cache and will be removed
569 // when this TokenLexer finishes lexing them.
570 Tokens = PP.cacheMacroExpandedTokens(this, ResultToks);
572 // The preprocessor cache of macro expanded tokens owns these tokens,not us.
577 /// \brief Checks if two tokens form wide string literal.
578 static bool isWideStringLiteralFromMacro(const Token &FirstTok,
579 const Token &SecondTok) {
580 return FirstTok.is(tok::identifier) &&
581 FirstTok.getIdentifierInfo()->isStr("L") && SecondTok.isLiteral() &&
582 SecondTok.stringifiedInMacro();
585 /// Lex - Lex and return a token from this macro stream.
586 bool TokenLexer::Lex(Token &Tok) {
587 // Lexing off the end of the macro, pop this macro off the expansion stack.
589 // If this is a macro (not a token stream), mark the macro enabled now
590 // that it is no longer being expanded.
591 if (Macro) Macro->EnableMacro();
594 Tok.setFlagValue(Token::StartOfLine , AtStartOfLine);
595 Tok.setFlagValue(Token::LeadingSpace, HasLeadingSpace || NextTokGetsSpace);
596 if (CurTokenIdx == 0)
597 Tok.setFlag(Token::LeadingEmptyMacro);
598 return PP.HandleEndOfTokenLexer(Tok);
601 SourceManager &SM = PP.getSourceManager();
603 // If this is the first token of the expanded result, we inherit spacing
605 bool isFirstToken = CurTokenIdx == 0;
607 // Get the next token to return.
608 Tok = Tokens[CurTokenIdx++];
610 bool TokenIsFromPaste = false;
612 // If this token is followed by a token paste (##) operator, paste the tokens!
613 // Note that ## is a normal token when not expanding a macro.
614 if (!isAtEnd() && Macro &&
615 (Tokens[CurTokenIdx].is(tok::hashhash) ||
616 // Special processing of L#x macros in -fms-compatibility mode.
617 // Microsoft compiler is able to form a wide string literal from
618 // 'L#macro_arg' construct in a function-like macro.
619 (PP.getLangOpts().MSVCCompat &&
620 isWideStringLiteralFromMacro(Tok, Tokens[CurTokenIdx])))) {
621 // When handling the microsoft /##/ extension, the final token is
622 // returned by pasteTokens, not the pasted token.
623 if (pasteTokens(Tok))
626 TokenIsFromPaste = true;
629 // The token's current location indicate where the token was lexed from. We
630 // need this information to compute the spelling of the token, but any
631 // diagnostics for the expanded token should appear as if they came from
632 // ExpansionLoc. Pull this information together into a new SourceLocation
633 // that captures all of this.
634 if (ExpandLocStart.isValid() && // Don't do this for token streams.
635 // Check that the token's location was not already set properly.
636 SM.isBeforeInSLocAddrSpace(Tok.getLocation(), MacroStartSLocOffset)) {
637 SourceLocation instLoc;
638 if (Tok.is(tok::comment)) {
639 instLoc = SM.createExpansionLoc(Tok.getLocation(),
644 instLoc = getExpansionLocForMacroDefLoc(Tok.getLocation());
647 Tok.setLocation(instLoc);
650 // If this is the first token, set the lexical properties of the token to
651 // match the lexical properties of the macro identifier.
653 Tok.setFlagValue(Token::StartOfLine , AtStartOfLine);
654 Tok.setFlagValue(Token::LeadingSpace, HasLeadingSpace);
656 // If this is not the first token, we may still need to pass through
657 // leading whitespace if we've expanded a macro.
658 if (AtStartOfLine) Tok.setFlag(Token::StartOfLine);
659 if (HasLeadingSpace) Tok.setFlag(Token::LeadingSpace);
661 AtStartOfLine = false;
662 HasLeadingSpace = false;
664 // Handle recursive expansion!
665 if (!Tok.isAnnotation() && Tok.getIdentifierInfo() != nullptr) {
666 // Change the kind of this identifier to the appropriate token kind, e.g.
667 // turning "for" into a keyword.
668 IdentifierInfo *II = Tok.getIdentifierInfo();
669 Tok.setKind(II->getTokenID());
671 // If this identifier was poisoned and from a paste, emit an error. This
672 // won't be handled by Preprocessor::HandleIdentifier because this is coming
673 // from a macro expansion.
674 if (II->isPoisoned() && TokenIsFromPaste) {
675 PP.HandlePoisonedIdentifier(Tok);
678 if (!DisableMacroExpansion && II->isHandleIdentifierCase())
679 return PP.HandleIdentifier(Tok);
682 // Otherwise, return a normal token.
686 bool TokenLexer::pasteTokens(Token &Tok) {
687 return pasteTokens(Tok, llvm::makeArrayRef(Tokens, NumTokens), CurTokenIdx);
690 /// LHSTok is the LHS of a ## operator, and CurTokenIdx is the ##
691 /// operator. Read the ## and RHS, and paste the LHS/RHS together. If there
692 /// are more ## after it, chomp them iteratively. Return the result as LHSTok.
693 /// If this returns true, the caller should immediately return the token.
694 bool TokenLexer::pasteTokens(Token &LHSTok, ArrayRef<Token> TokenStream,
695 unsigned int &CurIdx) {
696 assert(CurIdx > 0 && "## can not be the first token within tokens");
697 assert((TokenStream[CurIdx].is(tok::hashhash) ||
698 (PP.getLangOpts().MSVCCompat &&
699 isWideStringLiteralFromMacro(LHSTok, TokenStream[CurIdx]))) &&
700 "Token at this Index must be ## or part of the MSVC 'L "
701 "#macro-arg' pasting pair");
703 // MSVC: If previous token was pasted, this must be a recovery from an invalid
704 // paste operation. Ignore spaces before this token to mimic MSVC output.
705 // Required for generating valid UUID strings in some MS headers.
706 if (PP.getLangOpts().MicrosoftExt && (CurIdx >= 2) &&
707 TokenStream[CurIdx - 2].is(tok::hashhash))
708 LHSTok.clearFlag(Token::LeadingSpace);
710 SmallString<128> Buffer;
711 const char *ResultTokStrPtr = nullptr;
712 SourceLocation StartLoc = LHSTok.getLocation();
713 SourceLocation PasteOpLoc;
715 auto IsAtEnd = [&TokenStream, &CurIdx] {
716 return TokenStream.size() == CurIdx;
720 // Consume the ## operator if any.
721 PasteOpLoc = TokenStream[CurIdx].getLocation();
722 if (TokenStream[CurIdx].is(tok::hashhash))
724 assert(!IsAtEnd() && "No token on the RHS of a paste operator!");
726 // Get the RHS token.
727 const Token &RHS = TokenStream[CurIdx];
729 // Allocate space for the result token. This is guaranteed to be enough for
731 Buffer.resize(LHSTok.getLength() + RHS.getLength());
733 // Get the spelling of the LHS token in Buffer.
734 const char *BufPtr = &Buffer[0];
735 bool Invalid = false;
736 unsigned LHSLen = PP.getSpelling(LHSTok, BufPtr, &Invalid);
737 if (BufPtr != &Buffer[0]) // Really, we want the chars in Buffer!
738 memcpy(&Buffer[0], BufPtr, LHSLen);
742 BufPtr = Buffer.data() + LHSLen;
743 unsigned RHSLen = PP.getSpelling(RHS, BufPtr, &Invalid);
746 if (RHSLen && BufPtr != &Buffer[LHSLen])
747 // Really, we want the chars in Buffer!
748 memcpy(&Buffer[LHSLen], BufPtr, RHSLen);
750 // Trim excess space.
751 Buffer.resize(LHSLen+RHSLen);
753 // Plop the pasted result (including the trailing newline and null) into a
754 // scratch buffer where we can lex it.
756 ResultTokTmp.startToken();
758 // Claim that the tmp token is a string_literal so that we can get the
759 // character pointer back from CreateString in getLiteralData().
760 ResultTokTmp.setKind(tok::string_literal);
761 PP.CreateString(Buffer, ResultTokTmp);
762 SourceLocation ResultTokLoc = ResultTokTmp.getLocation();
763 ResultTokStrPtr = ResultTokTmp.getLiteralData();
765 // Lex the resultant pasted token into Result.
768 if (LHSTok.isAnyIdentifier() && RHS.isAnyIdentifier()) {
769 // Common paste case: identifier+identifier = identifier. Avoid creating
770 // a lexer and other overhead.
771 PP.IncrementPasteCounter(true);
773 Result.setKind(tok::raw_identifier);
774 Result.setRawIdentifierData(ResultTokStrPtr);
775 Result.setLocation(ResultTokLoc);
776 Result.setLength(LHSLen+RHSLen);
778 PP.IncrementPasteCounter(false);
780 assert(ResultTokLoc.isFileID() &&
781 "Should be a raw location into scratch buffer");
782 SourceManager &SourceMgr = PP.getSourceManager();
783 FileID LocFileID = SourceMgr.getFileID(ResultTokLoc);
785 bool Invalid = false;
786 const char *ScratchBufStart
787 = SourceMgr.getBufferData(LocFileID, &Invalid).data();
791 // Make a lexer to lex this string from. Lex just this one token.
792 // Make a lexer object so that we lex and expand the paste result.
793 Lexer TL(SourceMgr.getLocForStartOfFile(LocFileID),
794 PP.getLangOpts(), ScratchBufStart,
795 ResultTokStrPtr, ResultTokStrPtr+LHSLen+RHSLen);
797 // Lex a token in raw mode. This way it won't look up identifiers
798 // automatically, lexing off the end will return an eof token, and
799 // warnings are disabled. This returns true if the result token is the
801 bool isInvalid = !TL.LexFromRawLexer(Result);
803 // If we got an EOF token, we didn't form even ONE token. For example, we
804 // did "/ ## /" to get "//".
805 isInvalid |= Result.is(tok::eof);
807 // If pasting the two tokens didn't form a full new token, this is an
808 // error. This occurs with "x ## +" and other stuff. Return with LHSTok
809 // unmodified and with RHS as the next token to lex.
811 // Explicitly convert the token location to have proper expansion
812 // information so that the user knows where it came from.
813 SourceManager &SM = PP.getSourceManager();
815 SM.createExpansionLoc(PasteOpLoc, ExpandLocStart, ExpandLocEnd, 2);
817 // Test for the Microsoft extension of /##/ turning into // here on the
819 if (PP.getLangOpts().MicrosoftExt && LHSTok.is(tok::slash) &&
820 RHS.is(tok::slash)) {
821 HandleMicrosoftCommentPaste(LHSTok, Loc);
825 // Do not emit the error when preprocessing assembler code.
826 if (!PP.getLangOpts().AsmPreprocessor) {
827 // If we're in microsoft extensions mode, downgrade this from a hard
828 // error to an extension that defaults to an error. This allows
830 PP.Diag(Loc, PP.getLangOpts().MicrosoftExt ? diag::ext_pp_bad_paste_ms
831 : diag::err_pp_bad_paste)
835 // An error has occurred so exit loop.
839 // Turn ## into 'unknown' to avoid # ## # from looking like a paste
841 if (Result.is(tok::hashhash))
842 Result.setKind(tok::unknown);
845 // Transfer properties of the LHS over the Result.
846 Result.setFlagValue(Token::StartOfLine , LHSTok.isAtStartOfLine());
847 Result.setFlagValue(Token::LeadingSpace, LHSTok.hasLeadingSpace());
849 // Finally, replace LHS with the result, consume the RHS, and iterate.
852 } while (!IsAtEnd() && TokenStream[CurIdx].is(tok::hashhash));
854 SourceLocation EndLoc = TokenStream[CurIdx - 1].getLocation();
856 // The token's current location indicate where the token was lexed from. We
857 // need this information to compute the spelling of the token, but any
858 // diagnostics for the expanded token should appear as if the token was
859 // expanded from the full ## expression. Pull this information together into
860 // a new SourceLocation that captures all of this.
861 SourceManager &SM = PP.getSourceManager();
862 if (StartLoc.isFileID())
863 StartLoc = getExpansionLocForMacroDefLoc(StartLoc);
864 if (EndLoc.isFileID())
865 EndLoc = getExpansionLocForMacroDefLoc(EndLoc);
866 FileID MacroFID = SM.getFileID(MacroExpansionStart);
867 while (SM.getFileID(StartLoc) != MacroFID)
868 StartLoc = SM.getImmediateExpansionRange(StartLoc).first;
869 while (SM.getFileID(EndLoc) != MacroFID)
870 EndLoc = SM.getImmediateExpansionRange(EndLoc).second;
872 LHSTok.setLocation(SM.createExpansionLoc(LHSTok.getLocation(), StartLoc, EndLoc,
873 LHSTok.getLength()));
875 // Now that we got the result token, it will be subject to expansion. Since
876 // token pasting re-lexes the result token in raw mode, identifier information
877 // isn't looked up. As such, if the result is an identifier, look up id info.
878 if (LHSTok.is(tok::raw_identifier)) {
879 // Look up the identifier info for the token. We disabled identifier lookup
880 // by saying we're skipping contents, so we need to do this manually.
881 PP.LookUpIdentifierInfo(LHSTok);
886 /// isNextTokenLParen - If the next token lexed will pop this macro off the
887 /// expansion stack, return 2. If the next unexpanded token is a '(', return
888 /// 1, otherwise return 0.
889 unsigned TokenLexer::isNextTokenLParen() const {
893 return Tokens[CurTokenIdx].is(tok::l_paren);
896 /// isParsingPreprocessorDirective - Return true if we are in the middle of a
897 /// preprocessor directive.
898 bool TokenLexer::isParsingPreprocessorDirective() const {
899 return Tokens[NumTokens-1].is(tok::eod) && !isAtEnd();
902 /// HandleMicrosoftCommentPaste - In microsoft compatibility mode, /##/ pastes
903 /// together to form a comment that comments out everything in the current
904 /// macro, other active macros, and anything left on the current physical
905 /// source line of the expanded buffer. Handle this by returning the
906 /// first token on the next line.
907 void TokenLexer::HandleMicrosoftCommentPaste(Token &Tok, SourceLocation OpLoc) {
908 PP.Diag(OpLoc, diag::ext_comment_paste_microsoft);
910 // We 'comment out' the rest of this macro by just ignoring the rest of the
911 // tokens that have not been lexed yet, if any.
913 // Since this must be a macro, mark the macro enabled now that it is no longer
915 assert(Macro && "Token streams can't paste comments");
916 Macro->EnableMacro();
918 PP.HandleMicrosoftCommentPaste(Tok);
921 /// \brief If \arg loc is a file ID and points inside the current macro
922 /// definition, returns the appropriate source location pointing at the
923 /// macro expansion source location entry, otherwise it returns an invalid
926 TokenLexer::getExpansionLocForMacroDefLoc(SourceLocation loc) const {
927 assert(ExpandLocStart.isValid() && MacroExpansionStart.isValid() &&
928 "Not appropriate for token streams");
929 assert(loc.isValid() && loc.isFileID());
931 SourceManager &SM = PP.getSourceManager();
932 assert(SM.isInSLocAddrSpace(loc, MacroDefStart, MacroDefLength) &&
933 "Expected loc to come from the macro definition");
935 unsigned relativeOffset = 0;
936 SM.isInSLocAddrSpace(loc, MacroDefStart, MacroDefLength, &relativeOffset);
937 return MacroExpansionStart.getLocWithOffset(relativeOffset);
940 /// \brief Finds the tokens that are consecutive (from the same FileID)
941 /// creates a single SLocEntry, and assigns SourceLocations to each token that
942 /// point to that SLocEntry. e.g for
943 /// assert(foo == bar);
944 /// There will be a single SLocEntry for the "foo == bar" chunk and locations
945 /// for the 'foo', '==', 'bar' tokens will point inside that chunk.
947 /// \arg begin_tokens will be updated to a position past all the found
948 /// consecutive tokens.
949 static void updateConsecutiveMacroArgTokens(SourceManager &SM,
950 SourceLocation InstLoc,
951 Token *&begin_tokens,
952 Token * end_tokens) {
953 assert(begin_tokens < end_tokens);
955 SourceLocation FirstLoc = begin_tokens->getLocation();
956 SourceLocation CurLoc = FirstLoc;
958 // Compare the source location offset of tokens and group together tokens that
959 // are close, even if their locations point to different FileIDs. e.g.
961 // |bar | foo | cake | (3 tokens from 3 consecutive FileIDs)
963 // |bar foo cake| (one SLocEntry chunk for all tokens)
965 // we can perform this "merge" since the token's spelling location depends
966 // on the relative offset.
968 Token *NextTok = begin_tokens + 1;
969 for (; NextTok < end_tokens; ++NextTok) {
970 SourceLocation NextLoc = NextTok->getLocation();
971 if (CurLoc.isFileID() != NextLoc.isFileID())
972 break; // Token from different kind of FileID.
975 if (!SM.isInSameSLocAddrSpace(CurLoc, NextLoc, &RelOffs))
976 break; // Token from different local/loaded location.
977 // Check that token is not before the previous token or more than 50
978 // "characters" away.
979 if (RelOffs < 0 || RelOffs > 50)
982 if (CurLoc.isMacroID() && !SM.isWrittenInSameFile(CurLoc, NextLoc))
983 break; // Token from a different macro.
988 // For the consecutive tokens, find the length of the SLocEntry to contain
990 Token &LastConsecutiveTok = *(NextTok-1);
992 SM.isInSameSLocAddrSpace(FirstLoc, LastConsecutiveTok.getLocation(),
994 unsigned FullLength = LastRelOffs + LastConsecutiveTok.getLength();
996 // Create a macro expansion SLocEntry that will "contain" all of the tokens.
997 SourceLocation Expansion =
998 SM.createMacroArgExpansionLoc(FirstLoc, InstLoc,FullLength);
1000 // Change the location of the tokens from the spelling location to the new
1001 // expanded location.
1002 for (; begin_tokens < NextTok; ++begin_tokens) {
1003 Token &Tok = *begin_tokens;
1005 SM.isInSameSLocAddrSpace(FirstLoc, Tok.getLocation(), &RelOffs);
1006 Tok.setLocation(Expansion.getLocWithOffset(RelOffs));
1010 /// \brief Creates SLocEntries and updates the locations of macro argument
1011 /// tokens to their new expanded locations.
1013 /// \param ArgIdSpellLoc the location of the macro argument id inside the macro
1015 void TokenLexer::updateLocForMacroArgTokens(SourceLocation ArgIdSpellLoc,
1016 Token *begin_tokens,
1017 Token *end_tokens) {
1018 SourceManager &SM = PP.getSourceManager();
1020 SourceLocation InstLoc =
1021 getExpansionLocForMacroDefLoc(ArgIdSpellLoc);
1023 while (begin_tokens < end_tokens) {
1024 // If there's only one token just create a SLocEntry for it.
1025 if (end_tokens - begin_tokens == 1) {
1026 Token &Tok = *begin_tokens;
1027 Tok.setLocation(SM.createMacroArgExpansionLoc(Tok.getLocation(),
1033 updateConsecutiveMacroArgTokens(SM, InstLoc, begin_tokens, end_tokens);
1037 void TokenLexer::PropagateLineStartLeadingSpaceInfo(Token &Result) {
1038 AtStartOfLine = Result.isAtStartOfLine();
1039 HasLeadingSpace = Result.hasLeadingSpace();