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/SmallString.h" |
30 | #include "llvm/ADT/SmallVector.h" |
31 | #include "llvm/ADT/iterator_range.h" |
32 | #include <cassert> |
33 | #include <cstring> |
34 | #include <optional> |
35 | |
36 | using namespace clang; |
37 | |
38 | /// Create a TokenLexer for the specified macro with the specified actual |
39 | /// arguments. Note that this ctor takes ownership of the ActualArgs pointer. |
40 | void TokenLexer::Init(Token &Tok, SourceLocation ELEnd, MacroInfo *MI, |
41 | MacroArgs *Actuals) { |
42 | // If the client is reusing a TokenLexer, make sure to free any memory |
43 | // associated with it. |
44 | destroy(); |
45 | |
46 | Macro = MI; |
47 | ActualArgs = Actuals; |
48 | CurTokenIdx = 0; |
49 | |
50 | ExpandLocStart = Tok.getLocation(); |
51 | ExpandLocEnd = ELEnd; |
52 | AtStartOfLine = Tok.isAtStartOfLine(); |
53 | HasLeadingSpace = Tok.hasLeadingSpace(); |
54 | NextTokGetsSpace = false; |
55 | Tokens = &*Macro->tokens_begin(); |
56 | OwnsTokens = false; |
57 | DisableMacroExpansion = false; |
58 | IsReinject = false; |
59 | NumTokens = Macro->tokens_end()-Macro->tokens_begin(); |
60 | MacroExpansionStart = SourceLocation(); |
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 | |
118 | // Set HasLeadingSpace/AtStartOfLine so that the first token will be |
119 | // returned unmodified. |
120 | if (NumToks != 0) { |
121 | AtStartOfLine = TokArray[0].isAtStartOfLine(); |
122 | HasLeadingSpace = TokArray[0].hasLeadingSpace(); |
123 | } |
124 | } |
125 | |
126 | void TokenLexer::destroy() { |
127 | // If this was a function-like macro that actually uses its arguments, delete |
128 | // the expanded tokens. |
129 | if (OwnsTokens) { |
130 | delete [] Tokens; |
131 | Tokens = nullptr; |
132 | OwnsTokens = false; |
133 | } |
134 | |
135 | // TokenLexer owns its formal arguments. |
136 | if (ActualArgs) ActualArgs->destroy(PP); |
137 | } |
138 | |
139 | bool TokenLexer::MaybeRemoveCommaBeforeVaArgs( |
140 | SmallVectorImpl<Token> &ResultToks, bool HasPasteOperator, MacroInfo *Macro, |
141 | unsigned MacroArgNo, Preprocessor &PP) { |
142 | // Is the macro argument __VA_ARGS__? |
143 | if (!Macro->isVariadic() || MacroArgNo != Macro->getNumParams()-1) |
144 | return false; |
145 | |
146 | // In Microsoft-compatibility mode, a comma is removed in the expansion |
147 | // of " ... , __VA_ARGS__ " if __VA_ARGS__ is empty. This extension is |
148 | // not supported by gcc. |
149 | if (!HasPasteOperator && !PP.getLangOpts().MSVCCompat) |
150 | return false; |
151 | |
152 | // GCC removes the comma in the expansion of " ... , ## __VA_ARGS__ " if |
153 | // __VA_ARGS__ is empty, but not in strict C99 mode where there are no |
154 | // named arguments, where it remains. In all other modes, including C99 |
155 | // with GNU extensions, it is removed regardless of named arguments. |
156 | // Microsoft also appears to support this extension, unofficially. |
157 | if (PP.getLangOpts().C99 && !PP.getLangOpts().GNUMode |
158 | && Macro->getNumParams() < 2) |
159 | return false; |
160 | |
161 | // Is a comma available to be removed? |
162 | if (ResultToks.empty() || !ResultToks.back().is(K: tok::comma)) |
163 | return false; |
164 | |
165 | // Issue an extension diagnostic for the paste operator. |
166 | if (HasPasteOperator) |
167 | PP.Diag(Loc: ResultToks.back().getLocation(), DiagID: diag::ext_paste_comma); |
168 | |
169 | // Remove the comma. |
170 | ResultToks.pop_back(); |
171 | |
172 | if (!ResultToks.empty()) { |
173 | // If the comma was right after another paste (e.g. "X##,##__VA_ARGS__"), |
174 | // then removal of the comma should produce a placemarker token (in C99 |
175 | // terms) which we model by popping off the previous ##, giving us a plain |
176 | // "X" when __VA_ARGS__ is empty. |
177 | if (ResultToks.back().is(K: tok::hashhash)) |
178 | ResultToks.pop_back(); |
179 | |
180 | // Remember that this comma was elided. |
181 | ResultToks.back().setFlag(Token::CommaAfterElided); |
182 | } |
183 | |
184 | // Never add a space, even if the comma, ##, or arg had a space. |
185 | NextTokGetsSpace = false; |
186 | return true; |
187 | } |
188 | |
189 | void TokenLexer::stringifyVAOPTContents( |
190 | SmallVectorImpl<Token> &ResultToks, const VAOptExpansionContext &VCtx, |
191 | const SourceLocation VAOPTClosingParenLoc) { |
192 | const int NumToksPriorToVAOpt = VCtx.getNumberOfTokensPriorToVAOpt(); |
193 | const unsigned int NumVAOptTokens = ResultToks.size() - NumToksPriorToVAOpt; |
194 | Token *const VAOPTTokens = |
195 | NumVAOptTokens ? &ResultToks[NumToksPriorToVAOpt] : nullptr; |
196 | |
197 | SmallVector<Token, 64> ConcatenatedVAOPTResultToks; |
198 | // FIXME: Should we keep track within VCtx that we did or didnot |
199 | // encounter pasting - and only then perform this loop. |
200 | |
201 | // Perform token pasting (concatenation) prior to stringization. |
202 | for (unsigned int CurTokenIdx = 0; CurTokenIdx != NumVAOptTokens; |
203 | ++CurTokenIdx) { |
204 | if (VAOPTTokens[CurTokenIdx].is(K: tok::hashhash)) { |
205 | assert(CurTokenIdx != 0 && |
206 | "Can not have __VAOPT__ contents begin with a ##" ); |
207 | Token &LHS = VAOPTTokens[CurTokenIdx - 1]; |
208 | pasteTokens(LHSTok&: LHS, TokenStream: llvm::ArrayRef(VAOPTTokens, NumVAOptTokens), |
209 | CurIdx&: CurTokenIdx); |
210 | // Replace the token prior to the first ## in this iteration. |
211 | ConcatenatedVAOPTResultToks.back() = LHS; |
212 | if (CurTokenIdx == NumVAOptTokens) |
213 | break; |
214 | } |
215 | ConcatenatedVAOPTResultToks.push_back(Elt: VAOPTTokens[CurTokenIdx]); |
216 | } |
217 | |
218 | ConcatenatedVAOPTResultToks.push_back(Elt: VCtx.getEOFTok()); |
219 | // Get the SourceLocation that represents the start location within |
220 | // the macro definition that marks where this string is substituted |
221 | // into: i.e. the __VA_OPT__ and the ')' within the spelling of the |
222 | // macro definition, and use it to indicate that the stringified token |
223 | // was generated from that location. |
224 | const SourceLocation ExpansionLocStartWithinMacro = |
225 | getExpansionLocForMacroDefLoc(loc: VCtx.getVAOptLoc()); |
226 | const SourceLocation ExpansionLocEndWithinMacro = |
227 | getExpansionLocForMacroDefLoc(loc: VAOPTClosingParenLoc); |
228 | |
229 | Token StringifiedVAOPT = MacroArgs::StringifyArgument( |
230 | ArgToks: &ConcatenatedVAOPTResultToks[0], PP, Charify: VCtx.hasCharifyBefore() /*Charify*/, |
231 | ExpansionLocStart: ExpansionLocStartWithinMacro, ExpansionLocEnd: ExpansionLocEndWithinMacro); |
232 | |
233 | if (VCtx.getLeadingSpaceForStringifiedToken()) |
234 | StringifiedVAOPT.setFlag(Token::LeadingSpace); |
235 | |
236 | StringifiedVAOPT.setFlag(Token::StringifiedInMacro); |
237 | // Resize (shrink) the token stream to just capture this stringified token. |
238 | ResultToks.resize(N: NumToksPriorToVAOpt + 1); |
239 | ResultToks.back() = StringifiedVAOPT; |
240 | } |
241 | |
242 | /// Expand the arguments of a function-like macro so that we can quickly |
243 | /// return preexpanded tokens from Tokens. |
244 | void TokenLexer::ExpandFunctionArguments() { |
245 | SmallVector<Token, 128> ResultToks; |
246 | |
247 | // Loop through 'Tokens', expanding them into ResultToks. Keep |
248 | // track of whether we change anything. If not, no need to keep them. If so, |
249 | // we install the newly expanded sequence as the new 'Tokens' list. |
250 | bool MadeChange = false; |
251 | |
252 | std::optional<bool> CalledWithVariadicArguments; |
253 | |
254 | VAOptExpansionContext VCtx(PP); |
255 | |
256 | for (unsigned I = 0, E = NumTokens; I != E; ++I) { |
257 | const Token &CurTok = Tokens[I]; |
258 | // We don't want a space for the next token after a paste |
259 | // operator. In valid code, the token will get smooshed onto the |
260 | // preceding one anyway. In assembler-with-cpp mode, invalid |
261 | // pastes are allowed through: in this case, we do not want the |
262 | // extra whitespace to be added. For example, we want ". ## foo" |
263 | // -> ".foo" not ". foo". |
264 | if (I != 0 && !Tokens[I-1].is(K: tok::hashhash) && CurTok.hasLeadingSpace()) |
265 | NextTokGetsSpace = true; |
266 | |
267 | if (VCtx.isVAOptToken(T: CurTok)) { |
268 | MadeChange = true; |
269 | assert(Tokens[I + 1].is(tok::l_paren) && |
270 | "__VA_OPT__ must be followed by '('" ); |
271 | |
272 | ++I; // Skip the l_paren |
273 | VCtx.sawVAOptFollowedByOpeningParens(VAOptLoc: CurTok.getLocation(), |
274 | NumPriorTokens: ResultToks.size()); |
275 | |
276 | continue; |
277 | } |
278 | |
279 | // We have entered into the __VA_OPT__ context, so handle tokens |
280 | // appropriately. |
281 | if (VCtx.isInVAOpt()) { |
282 | // If we are about to process a token that is either an argument to |
283 | // __VA_OPT__ or its closing rparen, then: |
284 | // 1) If the token is the closing rparen that exits us out of __VA_OPT__, |
285 | // perform any necessary stringification or placemarker processing, |
286 | // and/or skip to the next token. |
287 | // 2) else if macro was invoked without variadic arguments skip this |
288 | // token. |
289 | // 3) else (macro was invoked with variadic arguments) process the token |
290 | // normally. |
291 | |
292 | if (Tokens[I].is(K: tok::l_paren)) |
293 | VCtx.sawOpeningParen(LParenLoc: Tokens[I].getLocation()); |
294 | // Continue skipping tokens within __VA_OPT__ if the macro was not |
295 | // called with variadic arguments, else let the rest of the loop handle |
296 | // this token. Note sawClosingParen() returns true only if the r_paren matches |
297 | // the closing r_paren of the __VA_OPT__. |
298 | if (!Tokens[I].is(K: tok::r_paren) || !VCtx.sawClosingParen()) { |
299 | // Lazily expand __VA_ARGS__ when we see the first __VA_OPT__. |
300 | if (!CalledWithVariadicArguments) { |
301 | CalledWithVariadicArguments = |
302 | ActualArgs->invokedWithVariadicArgument(MI: Macro, PP); |
303 | } |
304 | if (!*CalledWithVariadicArguments) { |
305 | // Skip this token. |
306 | continue; |
307 | } |
308 | // ... else the macro was called with variadic arguments, and we do not |
309 | // have a closing rparen - so process this token normally. |
310 | } else { |
311 | // Current token is the closing r_paren which marks the end of the |
312 | // __VA_OPT__ invocation, so handle any place-marker pasting (if |
313 | // empty) by removing hashhash either before (if exists) or after. And |
314 | // also stringify the entire contents if VAOPT was preceded by a hash, |
315 | // but do so only after any token concatenation that needs to occur |
316 | // within the contents of VAOPT. |
317 | |
318 | if (VCtx.hasStringifyOrCharifyBefore()) { |
319 | // Replace all the tokens just added from within VAOPT into a single |
320 | // stringified token. This requires token-pasting to eagerly occur |
321 | // within these tokens. If either the contents of VAOPT were empty |
322 | // or the macro wasn't called with any variadic arguments, the result |
323 | // is a token that represents an empty string. |
324 | stringifyVAOPTContents(ResultToks, VCtx, |
325 | /*ClosingParenLoc*/ VAOPTClosingParenLoc: Tokens[I].getLocation()); |
326 | |
327 | } else if (/*No tokens within VAOPT*/ |
328 | ResultToks.size() == VCtx.getNumberOfTokensPriorToVAOpt()) { |
329 | // Treat VAOPT as a placemarker token. Eat either the '##' before the |
330 | // RHS/VAOPT (if one exists, suggesting that the LHS (if any) to that |
331 | // hashhash was not a placemarker) or the '##' |
332 | // after VAOPT, but not both. |
333 | |
334 | if (ResultToks.size() && ResultToks.back().is(K: tok::hashhash)) { |
335 | ResultToks.pop_back(); |
336 | } else if ((I + 1 != E) && Tokens[I + 1].is(K: tok::hashhash)) { |
337 | ++I; // Skip the following hashhash. |
338 | } |
339 | } else { |
340 | // If there's a ## before the __VA_OPT__, we might have discovered |
341 | // that the __VA_OPT__ begins with a placeholder. We delay action on |
342 | // that to now to avoid messing up our stashed count of tokens before |
343 | // __VA_OPT__. |
344 | if (VCtx.beginsWithPlaceholder()) { |
345 | assert(VCtx.getNumberOfTokensPriorToVAOpt() > 0 && |
346 | ResultToks.size() >= VCtx.getNumberOfTokensPriorToVAOpt() && |
347 | ResultToks[VCtx.getNumberOfTokensPriorToVAOpt() - 1].is( |
348 | tok::hashhash) && |
349 | "no token paste before __VA_OPT__" ); |
350 | ResultToks.erase(CI: ResultToks.begin() + |
351 | VCtx.getNumberOfTokensPriorToVAOpt() - 1); |
352 | } |
353 | // If the expansion of __VA_OPT__ ends with a placeholder, eat any |
354 | // following '##' token. |
355 | if (VCtx.endsWithPlaceholder() && I + 1 != E && |
356 | Tokens[I + 1].is(K: tok::hashhash)) { |
357 | ++I; |
358 | } |
359 | } |
360 | VCtx.reset(); |
361 | // We processed __VA_OPT__'s closing paren (and the exit out of |
362 | // __VA_OPT__), so skip to the next token. |
363 | continue; |
364 | } |
365 | } |
366 | |
367 | // If we found the stringify operator, get the argument stringified. The |
368 | // preprocessor already verified that the following token is a macro |
369 | // parameter or __VA_OPT__ when the #define was lexed. |
370 | |
371 | if (CurTok.isOneOf(K1: tok::hash, K2: tok::hashat)) { |
372 | int ArgNo = Macro->getParameterNum(Arg: Tokens[I+1].getIdentifierInfo()); |
373 | assert((ArgNo != -1 || VCtx.isVAOptToken(Tokens[I + 1])) && |
374 | "Token following # is not an argument or __VA_OPT__!" ); |
375 | |
376 | if (ArgNo == -1) { |
377 | // Handle the __VA_OPT__ case. |
378 | VCtx.sawHashOrHashAtBefore(HasLeadingSpace: NextTokGetsSpace, |
379 | IsHashAt: CurTok.is(K: tok::hashat)); |
380 | continue; |
381 | } |
382 | // Else handle the simple argument case. |
383 | SourceLocation ExpansionLocStart = |
384 | getExpansionLocForMacroDefLoc(loc: CurTok.getLocation()); |
385 | SourceLocation ExpansionLocEnd = |
386 | getExpansionLocForMacroDefLoc(loc: Tokens[I+1].getLocation()); |
387 | |
388 | bool Charify = CurTok.is(K: tok::hashat); |
389 | const Token *UnexpArg = ActualArgs->getUnexpArgument(Arg: ArgNo); |
390 | Token Res = MacroArgs::StringifyArgument( |
391 | ArgToks: UnexpArg, PP, Charify, ExpansionLocStart, ExpansionLocEnd); |
392 | Res.setFlag(Token::StringifiedInMacro); |
393 | |
394 | // The stringified/charified string leading space flag gets set to match |
395 | // the #/#@ operator. |
396 | if (NextTokGetsSpace) |
397 | Res.setFlag(Token::LeadingSpace); |
398 | |
399 | ResultToks.push_back(Elt: Res); |
400 | MadeChange = true; |
401 | ++I; // Skip arg name. |
402 | NextTokGetsSpace = false; |
403 | continue; |
404 | } |
405 | |
406 | // Find out if there is a paste (##) operator before or after the token. |
407 | bool NonEmptyPasteBefore = |
408 | !ResultToks.empty() && ResultToks.back().is(K: tok::hashhash); |
409 | bool PasteBefore = I != 0 && Tokens[I-1].is(K: tok::hashhash); |
410 | bool PasteAfter = I+1 != E && Tokens[I+1].is(K: tok::hashhash); |
411 | bool RParenAfter = I+1 != E && Tokens[I+1].is(K: tok::r_paren); |
412 | |
413 | assert((!NonEmptyPasteBefore || PasteBefore || VCtx.isInVAOpt()) && |
414 | "unexpected ## in ResultToks" ); |
415 | |
416 | // Otherwise, if this is not an argument token, just add the token to the |
417 | // output buffer. |
418 | IdentifierInfo *II = CurTok.getIdentifierInfo(); |
419 | int ArgNo = II ? Macro->getParameterNum(Arg: II) : -1; |
420 | if (ArgNo == -1) { |
421 | // This isn't an argument, just add it. |
422 | ResultToks.push_back(Elt: CurTok); |
423 | |
424 | if (NextTokGetsSpace) { |
425 | ResultToks.back().setFlag(Token::LeadingSpace); |
426 | NextTokGetsSpace = false; |
427 | } else if (PasteBefore && !NonEmptyPasteBefore) |
428 | ResultToks.back().clearFlag(Flag: Token::LeadingSpace); |
429 | |
430 | continue; |
431 | } |
432 | |
433 | // An argument is expanded somehow, the result is different than the |
434 | // input. |
435 | MadeChange = true; |
436 | |
437 | // Otherwise, this is a use of the argument. |
438 | |
439 | // In Microsoft mode, remove the comma before __VA_ARGS__ to ensure there |
440 | // are no trailing commas if __VA_ARGS__ is empty. |
441 | if (!PasteBefore && ActualArgs->isVarargsElidedUse() && |
442 | MaybeRemoveCommaBeforeVaArgs(ResultToks, |
443 | /*HasPasteOperator=*/false, |
444 | Macro, MacroArgNo: ArgNo, PP)) |
445 | continue; |
446 | |
447 | // If it is not the LHS/RHS of a ## operator, we must pre-expand the |
448 | // argument and substitute the expanded tokens into the result. This is |
449 | // C99 6.10.3.1p1. |
450 | if (!PasteBefore && !PasteAfter) { |
451 | const Token *ResultArgToks; |
452 | |
453 | // Only preexpand the argument if it could possibly need it. This |
454 | // avoids some work in common cases. |
455 | const Token *ArgTok = ActualArgs->getUnexpArgument(Arg: ArgNo); |
456 | if (ActualArgs->ArgNeedsPreexpansion(ArgTok, PP)) |
457 | ResultArgToks = &ActualArgs->getPreExpArgument(Arg: ArgNo, PP)[0]; |
458 | else |
459 | ResultArgToks = ArgTok; // Use non-preexpanded tokens. |
460 | |
461 | // If the arg token expanded into anything, append it. |
462 | if (ResultArgToks->isNot(K: tok::eof)) { |
463 | size_t FirstResult = ResultToks.size(); |
464 | unsigned NumToks = MacroArgs::getArgLength(ArgPtr: ResultArgToks); |
465 | ResultToks.append(in_start: ResultArgToks, in_end: ResultArgToks+NumToks); |
466 | |
467 | // In Microsoft-compatibility mode, we follow MSVC's preprocessing |
468 | // behavior by not considering single commas from nested macro |
469 | // expansions as argument separators. Set a flag on the token so we can |
470 | // test for this later when the macro expansion is processed. |
471 | if (PP.getLangOpts().MSVCCompat && NumToks == 1 && |
472 | ResultToks.back().is(K: tok::comma)) |
473 | ResultToks.back().setFlag(Token::IgnoredComma); |
474 | |
475 | // If the '##' came from expanding an argument, turn it into 'unknown' |
476 | // to avoid pasting. |
477 | for (Token &Tok : llvm::drop_begin(RangeOrContainer&: ResultToks, N: FirstResult)) |
478 | if (Tok.is(K: tok::hashhash)) |
479 | Tok.setKind(tok::unknown); |
480 | |
481 | if(ExpandLocStart.isValid()) { |
482 | updateLocForMacroArgTokens(ArgIdSpellLoc: CurTok.getLocation(), |
483 | begin_tokens: ResultToks.begin()+FirstResult, |
484 | end_tokens: ResultToks.end()); |
485 | } |
486 | |
487 | // If any tokens were substituted from the argument, the whitespace |
488 | // before the first token should match the whitespace of the arg |
489 | // identifier. |
490 | ResultToks[FirstResult].setFlagValue(Flag: Token::LeadingSpace, |
491 | Val: NextTokGetsSpace); |
492 | ResultToks[FirstResult].setFlagValue(Flag: Token::StartOfLine, Val: false); |
493 | NextTokGetsSpace = false; |
494 | } else { |
495 | // We're creating a placeholder token. Usually this doesn't matter, |
496 | // but it can affect paste behavior when at the start or end of a |
497 | // __VA_OPT__. |
498 | if (NonEmptyPasteBefore) { |
499 | // We're imagining a placeholder token is inserted here. If this is |
500 | // the first token in a __VA_OPT__ after a ##, delete the ##. |
501 | assert(VCtx.isInVAOpt() && "should only happen inside a __VA_OPT__" ); |
502 | VCtx.hasPlaceholderAfterHashhashAtStart(); |
503 | } else if (RParenAfter) |
504 | VCtx.hasPlaceholderBeforeRParen(); |
505 | } |
506 | continue; |
507 | } |
508 | |
509 | // Okay, we have a token that is either the LHS or RHS of a paste (##) |
510 | // argument. It gets substituted as its non-pre-expanded tokens. |
511 | const Token *ArgToks = ActualArgs->getUnexpArgument(Arg: ArgNo); |
512 | unsigned NumToks = MacroArgs::getArgLength(ArgPtr: ArgToks); |
513 | if (NumToks) { // Not an empty argument? |
514 | bool VaArgsPseudoPaste = false; |
515 | // If this is the GNU ", ## __VA_ARGS__" extension, and we just learned |
516 | // that __VA_ARGS__ expands to multiple tokens, avoid a pasting error when |
517 | // the expander tries to paste ',' with the first token of the __VA_ARGS__ |
518 | // expansion. |
519 | if (NonEmptyPasteBefore && ResultToks.size() >= 2 && |
520 | ResultToks[ResultToks.size()-2].is(K: tok::comma) && |
521 | (unsigned)ArgNo == Macro->getNumParams()-1 && |
522 | Macro->isVariadic()) { |
523 | VaArgsPseudoPaste = true; |
524 | // Remove the paste operator, report use of the extension. |
525 | PP.Diag(Loc: ResultToks.pop_back_val().getLocation(), DiagID: diag::ext_paste_comma); |
526 | } |
527 | |
528 | ResultToks.append(in_start: ArgToks, in_end: ArgToks+NumToks); |
529 | |
530 | // If the '##' came from expanding an argument, turn it into 'unknown' |
531 | // to avoid pasting. |
532 | for (Token &Tok : llvm::make_range(x: ResultToks.end() - NumToks, |
533 | y: ResultToks.end())) { |
534 | if (Tok.is(K: tok::hashhash)) |
535 | Tok.setKind(tok::unknown); |
536 | } |
537 | |
538 | if (ExpandLocStart.isValid()) { |
539 | updateLocForMacroArgTokens(ArgIdSpellLoc: CurTok.getLocation(), |
540 | begin_tokens: ResultToks.end()-NumToks, end_tokens: ResultToks.end()); |
541 | } |
542 | |
543 | // Transfer the leading whitespace information from the token |
544 | // (the macro argument) onto the first token of the |
545 | // expansion. Note that we don't do this for the GNU |
546 | // pseudo-paste extension ", ## __VA_ARGS__". |
547 | if (!VaArgsPseudoPaste) { |
548 | ResultToks[ResultToks.size() - NumToks].setFlagValue(Flag: Token::StartOfLine, |
549 | Val: false); |
550 | ResultToks[ResultToks.size() - NumToks].setFlagValue( |
551 | Flag: Token::LeadingSpace, Val: NextTokGetsSpace); |
552 | } |
553 | |
554 | NextTokGetsSpace = false; |
555 | continue; |
556 | } |
557 | |
558 | // If an empty argument is on the LHS or RHS of a paste, the standard (C99 |
559 | // 6.10.3.3p2,3) calls for a bunch of placemarker stuff to occur. We |
560 | // implement this by eating ## operators when a LHS or RHS expands to |
561 | // empty. |
562 | if (PasteAfter) { |
563 | // Discard the argument token and skip (don't copy to the expansion |
564 | // buffer) the paste operator after it. |
565 | ++I; |
566 | continue; |
567 | } |
568 | |
569 | if (RParenAfter && !NonEmptyPasteBefore) |
570 | VCtx.hasPlaceholderBeforeRParen(); |
571 | |
572 | // If this is on the RHS of a paste operator, we've already copied the |
573 | // paste operator to the ResultToks list, unless the LHS was empty too. |
574 | // Remove it. |
575 | assert(PasteBefore); |
576 | if (NonEmptyPasteBefore) { |
577 | assert(ResultToks.back().is(tok::hashhash)); |
578 | // Do not remove the paste operator if it is the one before __VA_OPT__ |
579 | // (and we are still processing tokens within VA_OPT). We handle the case |
580 | // of removing the paste operator if __VA_OPT__ reduces to the notional |
581 | // placemarker above when we encounter the closing paren of VA_OPT. |
582 | if (!VCtx.isInVAOpt() || |
583 | ResultToks.size() > VCtx.getNumberOfTokensPriorToVAOpt()) |
584 | ResultToks.pop_back(); |
585 | else |
586 | VCtx.hasPlaceholderAfterHashhashAtStart(); |
587 | } |
588 | |
589 | // If this is the __VA_ARGS__ token, and if the argument wasn't provided, |
590 | // and if the macro had at least one real argument, and if the token before |
591 | // the ## was a comma, remove the comma. This is a GCC extension which is |
592 | // disabled when using -std=c99. |
593 | if (ActualArgs->isVarargsElidedUse()) |
594 | MaybeRemoveCommaBeforeVaArgs(ResultToks, |
595 | /*HasPasteOperator=*/true, |
596 | Macro, MacroArgNo: ArgNo, PP); |
597 | } |
598 | |
599 | // If anything changed, install this as the new Tokens list. |
600 | if (MadeChange) { |
601 | assert(!OwnsTokens && "This would leak if we already own the token list" ); |
602 | // This is deleted in the dtor. |
603 | NumTokens = ResultToks.size(); |
604 | // The tokens will be added to Preprocessor's cache and will be removed |
605 | // when this TokenLexer finishes lexing them. |
606 | Tokens = PP.cacheMacroExpandedTokens(tokLexer: this, tokens: ResultToks); |
607 | |
608 | // The preprocessor cache of macro expanded tokens owns these tokens,not us. |
609 | OwnsTokens = false; |
610 | } |
611 | } |
612 | |
613 | /// Checks if two tokens form wide string literal. |
614 | static bool isWideStringLiteralFromMacro(const Token &FirstTok, |
615 | const Token &SecondTok) { |
616 | return FirstTok.is(K: tok::identifier) && |
617 | FirstTok.getIdentifierInfo()->isStr(Str: "L" ) && SecondTok.isLiteral() && |
618 | SecondTok.stringifiedInMacro(); |
619 | } |
620 | |
621 | /// Lex - Lex and return a token from this macro stream. |
622 | bool TokenLexer::Lex(Token &Tok) { |
623 | // Lexing off the end of the macro, pop this macro off the expansion stack. |
624 | if (isAtEnd()) { |
625 | // If this is a macro (not a token stream), mark the macro enabled now |
626 | // that it is no longer being expanded. |
627 | if (Macro) Macro->EnableMacro(); |
628 | |
629 | Tok.startToken(); |
630 | Tok.setFlagValue(Flag: Token::StartOfLine , Val: AtStartOfLine); |
631 | Tok.setFlagValue(Flag: Token::LeadingSpace, Val: HasLeadingSpace || NextTokGetsSpace); |
632 | if (CurTokenIdx == 0) |
633 | Tok.setFlag(Token::LeadingEmptyMacro); |
634 | return PP.HandleEndOfTokenLexer(Result&: Tok); |
635 | } |
636 | |
637 | SourceManager &SM = PP.getSourceManager(); |
638 | |
639 | // If this is the first token of the expanded result, we inherit spacing |
640 | // properties later. |
641 | bool isFirstToken = CurTokenIdx == 0; |
642 | |
643 | // Get the next token to return. |
644 | Tok = Tokens[CurTokenIdx++]; |
645 | if (IsReinject) |
646 | Tok.setFlag(Token::IsReinjected); |
647 | |
648 | bool TokenIsFromPaste = false; |
649 | |
650 | // If this token is followed by a token paste (##) operator, paste the tokens! |
651 | // Note that ## is a normal token when not expanding a macro. |
652 | if (!isAtEnd() && Macro && |
653 | (Tokens[CurTokenIdx].is(K: tok::hashhash) || |
654 | // Special processing of L#x macros in -fms-compatibility mode. |
655 | // Microsoft compiler is able to form a wide string literal from |
656 | // 'L#macro_arg' construct in a function-like macro. |
657 | (PP.getLangOpts().MSVCCompat && |
658 | isWideStringLiteralFromMacro(FirstTok: Tok, SecondTok: Tokens[CurTokenIdx])))) { |
659 | // When handling the microsoft /##/ extension, the final token is |
660 | // returned by pasteTokens, not the pasted token. |
661 | if (pasteTokens(Tok)) |
662 | return true; |
663 | |
664 | TokenIsFromPaste = true; |
665 | } |
666 | |
667 | // The token's current location indicate where the token was lexed from. We |
668 | // need this information to compute the spelling of the token, but any |
669 | // diagnostics for the expanded token should appear as if they came from |
670 | // ExpansionLoc. Pull this information together into a new SourceLocation |
671 | // that captures all of this. |
672 | if (ExpandLocStart.isValid() && // Don't do this for token streams. |
673 | // Check that the token's location was not already set properly. |
674 | SM.isBeforeInSLocAddrSpace(LHS: Tok.getLocation(), RHS: MacroStartSLocOffset)) { |
675 | SourceLocation instLoc; |
676 | if (Tok.is(K: tok::comment)) { |
677 | instLoc = SM.createExpansionLoc(SpellingLoc: Tok.getLocation(), |
678 | ExpansionLocStart: ExpandLocStart, |
679 | ExpansionLocEnd: ExpandLocEnd, |
680 | Length: Tok.getLength()); |
681 | } else { |
682 | instLoc = getExpansionLocForMacroDefLoc(loc: Tok.getLocation()); |
683 | } |
684 | |
685 | Tok.setLocation(instLoc); |
686 | } |
687 | |
688 | // If this is the first token, set the lexical properties of the token to |
689 | // match the lexical properties of the macro identifier. |
690 | if (isFirstToken) { |
691 | Tok.setFlagValue(Flag: Token::StartOfLine , Val: AtStartOfLine); |
692 | Tok.setFlagValue(Flag: Token::LeadingSpace, Val: HasLeadingSpace); |
693 | } else { |
694 | // If this is not the first token, we may still need to pass through |
695 | // leading whitespace if we've expanded a macro. |
696 | if (AtStartOfLine) Tok.setFlag(Token::StartOfLine); |
697 | if (HasLeadingSpace) Tok.setFlag(Token::LeadingSpace); |
698 | } |
699 | AtStartOfLine = false; |
700 | HasLeadingSpace = false; |
701 | |
702 | // Handle recursive expansion! |
703 | if (!Tok.isAnnotation() && Tok.getIdentifierInfo() != nullptr) { |
704 | // Change the kind of this identifier to the appropriate token kind, e.g. |
705 | // turning "for" into a keyword. |
706 | IdentifierInfo *II = Tok.getIdentifierInfo(); |
707 | Tok.setKind(II->getTokenID()); |
708 | |
709 | // If this identifier was poisoned and from a paste, emit an error. This |
710 | // won't be handled by Preprocessor::HandleIdentifier because this is coming |
711 | // from a macro expansion. |
712 | if (II->isPoisoned() && TokenIsFromPaste) { |
713 | PP.HandlePoisonedIdentifier(Identifier&: Tok); |
714 | } |
715 | |
716 | if (!DisableMacroExpansion && II->isHandleIdentifierCase()) |
717 | return PP.HandleIdentifier(Identifier&: Tok); |
718 | } |
719 | |
720 | // Otherwise, return a normal token. |
721 | return true; |
722 | } |
723 | |
724 | bool TokenLexer::pasteTokens(Token &Tok) { |
725 | return pasteTokens(LHSTok&: Tok, TokenStream: llvm::ArrayRef(Tokens, NumTokens), CurIdx&: CurTokenIdx); |
726 | } |
727 | |
728 | /// LHSTok is the LHS of a ## operator, and CurTokenIdx is the ## |
729 | /// operator. Read the ## and RHS, and paste the LHS/RHS together. If there |
730 | /// are more ## after it, chomp them iteratively. Return the result as LHSTok. |
731 | /// If this returns true, the caller should immediately return the token. |
732 | bool TokenLexer::pasteTokens(Token &LHSTok, ArrayRef<Token> TokenStream, |
733 | unsigned int &CurIdx) { |
734 | assert(CurIdx > 0 && "## can not be the first token within tokens" ); |
735 | assert((TokenStream[CurIdx].is(tok::hashhash) || |
736 | (PP.getLangOpts().MSVCCompat && |
737 | isWideStringLiteralFromMacro(LHSTok, TokenStream[CurIdx]))) && |
738 | "Token at this Index must be ## or part of the MSVC 'L " |
739 | "#macro-arg' pasting pair" ); |
740 | |
741 | // MSVC: If previous token was pasted, this must be a recovery from an invalid |
742 | // paste operation. Ignore spaces before this token to mimic MSVC output. |
743 | // Required for generating valid UUID strings in some MS headers. |
744 | if (PP.getLangOpts().MicrosoftExt && (CurIdx >= 2) && |
745 | TokenStream[CurIdx - 2].is(K: tok::hashhash)) |
746 | LHSTok.clearFlag(Flag: Token::LeadingSpace); |
747 | |
748 | SmallString<128> Buffer; |
749 | const char *ResultTokStrPtr = nullptr; |
750 | SourceLocation StartLoc = LHSTok.getLocation(); |
751 | SourceLocation PasteOpLoc; |
752 | |
753 | auto IsAtEnd = [&TokenStream, &CurIdx] { |
754 | return TokenStream.size() == CurIdx; |
755 | }; |
756 | |
757 | do { |
758 | // Consume the ## operator if any. |
759 | PasteOpLoc = TokenStream[CurIdx].getLocation(); |
760 | if (TokenStream[CurIdx].is(K: tok::hashhash)) |
761 | ++CurIdx; |
762 | assert(!IsAtEnd() && "No token on the RHS of a paste operator!" ); |
763 | |
764 | // Get the RHS token. |
765 | const Token &RHS = TokenStream[CurIdx]; |
766 | |
767 | // Allocate space for the result token. This is guaranteed to be enough for |
768 | // the two tokens. |
769 | Buffer.resize(N: LHSTok.getLength() + RHS.getLength()); |
770 | |
771 | // Get the spelling of the LHS token in Buffer. |
772 | const char *BufPtr = &Buffer[0]; |
773 | bool Invalid = false; |
774 | unsigned LHSLen = PP.getSpelling(Tok: LHSTok, Buffer&: BufPtr, Invalid: &Invalid); |
775 | if (BufPtr != &Buffer[0]) // Really, we want the chars in Buffer! |
776 | memcpy(dest: &Buffer[0], src: BufPtr, n: LHSLen); |
777 | if (Invalid) |
778 | return true; |
779 | |
780 | BufPtr = Buffer.data() + LHSLen; |
781 | unsigned RHSLen = PP.getSpelling(Tok: RHS, Buffer&: BufPtr, Invalid: &Invalid); |
782 | if (Invalid) |
783 | return true; |
784 | if (RHSLen && BufPtr != &Buffer[LHSLen]) |
785 | // Really, we want the chars in Buffer! |
786 | memcpy(dest: &Buffer[LHSLen], src: BufPtr, n: RHSLen); |
787 | |
788 | // Trim excess space. |
789 | Buffer.resize(N: LHSLen+RHSLen); |
790 | |
791 | // Plop the pasted result (including the trailing newline and null) into a |
792 | // scratch buffer where we can lex it. |
793 | Token ResultTokTmp; |
794 | ResultTokTmp.startToken(); |
795 | |
796 | // Claim that the tmp token is a string_literal so that we can get the |
797 | // character pointer back from CreateString in getLiteralData(). |
798 | ResultTokTmp.setKind(tok::string_literal); |
799 | PP.CreateString(Str: Buffer, Tok&: ResultTokTmp); |
800 | SourceLocation ResultTokLoc = ResultTokTmp.getLocation(); |
801 | ResultTokStrPtr = ResultTokTmp.getLiteralData(); |
802 | |
803 | // Lex the resultant pasted token into Result. |
804 | Token Result; |
805 | |
806 | if (LHSTok.isAnyIdentifier() && RHS.isAnyIdentifier()) { |
807 | // Common paste case: identifier+identifier = identifier. Avoid creating |
808 | // a lexer and other overhead. |
809 | PP.IncrementPasteCounter(isFast: true); |
810 | Result.startToken(); |
811 | Result.setKind(tok::raw_identifier); |
812 | Result.setRawIdentifierData(ResultTokStrPtr); |
813 | Result.setLocation(ResultTokLoc); |
814 | Result.setLength(LHSLen+RHSLen); |
815 | } else { |
816 | PP.IncrementPasteCounter(isFast: false); |
817 | |
818 | assert(ResultTokLoc.isFileID() && |
819 | "Should be a raw location into scratch buffer" ); |
820 | SourceManager &SourceMgr = PP.getSourceManager(); |
821 | FileID LocFileID = SourceMgr.getFileID(SpellingLoc: ResultTokLoc); |
822 | |
823 | bool Invalid = false; |
824 | const char *ScratchBufStart |
825 | = SourceMgr.getBufferData(FID: LocFileID, Invalid: &Invalid).data(); |
826 | if (Invalid) |
827 | return false; |
828 | |
829 | // Make a lexer to lex this string from. Lex just this one token. |
830 | // Make a lexer object so that we lex and expand the paste result. |
831 | Lexer TL(SourceMgr.getLocForStartOfFile(FID: LocFileID), |
832 | PP.getLangOpts(), ScratchBufStart, |
833 | ResultTokStrPtr, ResultTokStrPtr+LHSLen+RHSLen); |
834 | |
835 | // Lex a token in raw mode. This way it won't look up identifiers |
836 | // automatically, lexing off the end will return an eof token, and |
837 | // warnings are disabled. This returns true if the result token is the |
838 | // entire buffer. |
839 | bool isInvalid = !TL.LexFromRawLexer(Result); |
840 | |
841 | // If we got an EOF token, we didn't form even ONE token. For example, we |
842 | // did "/ ## /" to get "//". |
843 | isInvalid |= Result.is(K: tok::eof); |
844 | |
845 | // If pasting the two tokens didn't form a full new token, this is an |
846 | // error. This occurs with "x ## +" and other stuff. Return with LHSTok |
847 | // unmodified and with RHS as the next token to lex. |
848 | if (isInvalid) { |
849 | // Explicitly convert the token location to have proper expansion |
850 | // information so that the user knows where it came from. |
851 | SourceManager &SM = PP.getSourceManager(); |
852 | SourceLocation Loc = |
853 | SM.createExpansionLoc(SpellingLoc: PasteOpLoc, ExpansionLocStart: ExpandLocStart, ExpansionLocEnd: ExpandLocEnd, Length: 2); |
854 | |
855 | // Test for the Microsoft extension of /##/ turning into // here on the |
856 | // error path. |
857 | if (PP.getLangOpts().MicrosoftExt && LHSTok.is(K: tok::slash) && |
858 | RHS.is(K: tok::slash)) { |
859 | HandleMicrosoftCommentPaste(Tok&: LHSTok, OpLoc: Loc); |
860 | return true; |
861 | } |
862 | |
863 | // Do not emit the error when preprocessing assembler code. |
864 | if (!PP.getLangOpts().AsmPreprocessor) { |
865 | // If we're in microsoft extensions mode, downgrade this from a hard |
866 | // error to an extension that defaults to an error. This allows |
867 | // disabling it. |
868 | PP.Diag(Loc, DiagID: PP.getLangOpts().MicrosoftExt ? diag::ext_pp_bad_paste_ms |
869 | : diag::err_pp_bad_paste) |
870 | << Buffer; |
871 | } |
872 | |
873 | // An error has occurred so exit loop. |
874 | break; |
875 | } |
876 | |
877 | // Turn ## into 'unknown' to avoid # ## # from looking like a paste |
878 | // operator. |
879 | if (Result.is(K: tok::hashhash)) |
880 | Result.setKind(tok::unknown); |
881 | } |
882 | |
883 | // Transfer properties of the LHS over the Result. |
884 | Result.setFlagValue(Flag: Token::StartOfLine , Val: LHSTok.isAtStartOfLine()); |
885 | Result.setFlagValue(Flag: Token::LeadingSpace, Val: LHSTok.hasLeadingSpace()); |
886 | |
887 | // Finally, replace LHS with the result, consume the RHS, and iterate. |
888 | ++CurIdx; |
889 | LHSTok = Result; |
890 | } while (!IsAtEnd() && TokenStream[CurIdx].is(K: tok::hashhash)); |
891 | |
892 | SourceLocation EndLoc = TokenStream[CurIdx - 1].getLocation(); |
893 | |
894 | // The token's current location indicate where the token was lexed from. We |
895 | // need this information to compute the spelling of the token, but any |
896 | // diagnostics for the expanded token should appear as if the token was |
897 | // expanded from the full ## expression. Pull this information together into |
898 | // a new SourceLocation that captures all of this. |
899 | SourceManager &SM = PP.getSourceManager(); |
900 | if (StartLoc.isFileID()) |
901 | StartLoc = getExpansionLocForMacroDefLoc(loc: StartLoc); |
902 | if (EndLoc.isFileID()) |
903 | EndLoc = getExpansionLocForMacroDefLoc(loc: EndLoc); |
904 | FileID MacroFID = SM.getFileID(SpellingLoc: MacroExpansionStart); |
905 | while (SM.getFileID(SpellingLoc: StartLoc) != MacroFID) |
906 | StartLoc = SM.getImmediateExpansionRange(Loc: StartLoc).getBegin(); |
907 | while (SM.getFileID(SpellingLoc: EndLoc) != MacroFID) |
908 | EndLoc = SM.getImmediateExpansionRange(Loc: EndLoc).getEnd(); |
909 | |
910 | LHSTok.setLocation(SM.createExpansionLoc(SpellingLoc: LHSTok.getLocation(), ExpansionLocStart: StartLoc, ExpansionLocEnd: EndLoc, |
911 | Length: LHSTok.getLength())); |
912 | |
913 | // Now that we got the result token, it will be subject to expansion. Since |
914 | // token pasting re-lexes the result token in raw mode, identifier information |
915 | // isn't looked up. As such, if the result is an identifier, look up id info. |
916 | if (LHSTok.is(K: tok::raw_identifier)) { |
917 | // Look up the identifier info for the token. We disabled identifier lookup |
918 | // by saying we're skipping contents, so we need to do this manually. |
919 | PP.LookUpIdentifierInfo(Identifier&: LHSTok); |
920 | } |
921 | return false; |
922 | } |
923 | |
924 | /// isNextTokenLParen - If the next token lexed will pop this macro off the |
925 | /// expansion stack, return 2. If the next unexpanded token is a '(', return |
926 | /// 1, otherwise return 0. |
927 | unsigned TokenLexer::isNextTokenLParen() const { |
928 | // Out of tokens? |
929 | if (isAtEnd()) |
930 | return 2; |
931 | return Tokens[CurTokenIdx].is(K: tok::l_paren); |
932 | } |
933 | |
934 | /// isParsingPreprocessorDirective - Return true if we are in the middle of a |
935 | /// preprocessor directive. |
936 | bool TokenLexer::isParsingPreprocessorDirective() const { |
937 | return Tokens[NumTokens-1].is(K: tok::eod) && !isAtEnd(); |
938 | } |
939 | |
940 | /// HandleMicrosoftCommentPaste - In microsoft compatibility mode, /##/ pastes |
941 | /// together to form a comment that comments out everything in the current |
942 | /// macro, other active macros, and anything left on the current physical |
943 | /// source line of the expanded buffer. Handle this by returning the |
944 | /// first token on the next line. |
945 | void TokenLexer::HandleMicrosoftCommentPaste(Token &Tok, SourceLocation OpLoc) { |
946 | PP.Diag(Loc: OpLoc, DiagID: diag::ext_comment_paste_microsoft); |
947 | |
948 | // We 'comment out' the rest of this macro by just ignoring the rest of the |
949 | // tokens that have not been lexed yet, if any. |
950 | |
951 | // Since this must be a macro, mark the macro enabled now that it is no longer |
952 | // being expanded. |
953 | assert(Macro && "Token streams can't paste comments" ); |
954 | Macro->EnableMacro(); |
955 | |
956 | PP.HandleMicrosoftCommentPaste(Tok); |
957 | } |
958 | |
959 | /// If \arg loc is a file ID and points inside the current macro |
960 | /// definition, returns the appropriate source location pointing at the |
961 | /// macro expansion source location entry, otherwise it returns an invalid |
962 | /// SourceLocation. |
963 | SourceLocation |
964 | TokenLexer::getExpansionLocForMacroDefLoc(SourceLocation loc) const { |
965 | assert(ExpandLocStart.isValid() && MacroExpansionStart.isValid() && |
966 | "Not appropriate for token streams" ); |
967 | assert(loc.isValid() && loc.isFileID()); |
968 | |
969 | SourceManager &SM = PP.getSourceManager(); |
970 | assert(SM.isInSLocAddrSpace(loc, MacroDefStart, MacroDefLength) && |
971 | "Expected loc to come from the macro definition" ); |
972 | |
973 | SourceLocation::UIntTy relativeOffset = 0; |
974 | SM.isInSLocAddrSpace(Loc: loc, Start: MacroDefStart, Length: MacroDefLength, RelativeOffset: &relativeOffset); |
975 | return MacroExpansionStart.getLocWithOffset(Offset: relativeOffset); |
976 | } |
977 | |
978 | /// Finds the tokens that are consecutive (from the same FileID) |
979 | /// creates a single SLocEntry, and assigns SourceLocations to each token that |
980 | /// point to that SLocEntry. e.g for |
981 | /// assert(foo == bar); |
982 | /// There will be a single SLocEntry for the "foo == bar" chunk and locations |
983 | /// for the 'foo', '==', 'bar' tokens will point inside that chunk. |
984 | /// |
985 | /// \arg begin_tokens will be updated to a position past all the found |
986 | /// consecutive tokens. |
987 | static void updateConsecutiveMacroArgTokens(SourceManager &SM, |
988 | SourceLocation ExpandLoc, |
989 | Token *&begin_tokens, |
990 | Token * end_tokens) { |
991 | assert(begin_tokens + 1 < end_tokens); |
992 | SourceLocation BeginLoc = begin_tokens->getLocation(); |
993 | llvm::MutableArrayRef<Token> All(begin_tokens, end_tokens); |
994 | llvm::MutableArrayRef<Token> Partition; |
995 | |
996 | auto NearLast = [&, Last = BeginLoc](SourceLocation Loc) mutable { |
997 | // The maximum distance between two consecutive tokens in a partition. |
998 | // This is an important trick to avoid using too much SourceLocation address |
999 | // space! |
1000 | static constexpr SourceLocation::IntTy MaxDistance = 50; |
1001 | auto Distance = Loc.getRawEncoding() - Last.getRawEncoding(); |
1002 | Last = Loc; |
1003 | return Distance <= MaxDistance; |
1004 | }; |
1005 | |
1006 | // Partition the tokens by their FileID. |
1007 | // This is a hot function, and calling getFileID can be expensive, the |
1008 | // implementation is optimized by reducing the number of getFileID. |
1009 | if (BeginLoc.isFileID()) { |
1010 | // Consecutive tokens not written in macros must be from the same file. |
1011 | // (Neither #include nor eof can occur inside a macro argument.) |
1012 | Partition = All.take_while(Pred: [&](const Token &T) { |
1013 | return T.getLocation().isFileID() && NearLast(T.getLocation()); |
1014 | }); |
1015 | } else { |
1016 | // Call getFileID once to calculate the bounds, and use the cheaper |
1017 | // sourcelocation-against-bounds comparison. |
1018 | FileID BeginFID = SM.getFileID(SpellingLoc: BeginLoc); |
1019 | SourceLocation Limit = |
1020 | SM.getComposedLoc(FID: BeginFID, Offset: SM.getFileIDSize(FID: BeginFID)); |
1021 | Partition = All.take_while(Pred: [&](const Token &T) { |
1022 | // NOTE: the Limit is included! The lexer recovery only ever inserts a |
1023 | // single token past the end of the FileID, specifically the ) when a |
1024 | // macro-arg containing a comma should be guarded by parentheses. |
1025 | // |
1026 | // It is safe to include the Limit here because SourceManager allocates |
1027 | // FileSize + 1 for each SLocEntry. |
1028 | // |
1029 | // See https://github.com/llvm/llvm-project/issues/60722. |
1030 | return T.getLocation() >= BeginLoc && T.getLocation() <= Limit |
1031 | && NearLast(T.getLocation()); |
1032 | }); |
1033 | } |
1034 | assert(!Partition.empty()); |
1035 | |
1036 | // For the consecutive tokens, find the length of the SLocEntry to contain |
1037 | // all of them. |
1038 | SourceLocation::UIntTy FullLength = |
1039 | Partition.back().getEndLoc().getRawEncoding() - |
1040 | Partition.front().getLocation().getRawEncoding(); |
1041 | // Create a macro expansion SLocEntry that will "contain" all of the tokens. |
1042 | SourceLocation Expansion = |
1043 | SM.createMacroArgExpansionLoc(SpellingLoc: BeginLoc, ExpansionLoc: ExpandLoc, Length: FullLength); |
1044 | |
1045 | #ifdef EXPENSIVE_CHECKS |
1046 | assert(llvm::all_of(Partition.drop_front(), |
1047 | [&SM, ID = SM.getFileID(Partition.front().getLocation())]( |
1048 | const Token &T) { |
1049 | return ID == SM.getFileID(T.getLocation()); |
1050 | }) && |
1051 | "Must have the same FIleID!" ); |
1052 | #endif |
1053 | // Change the location of the tokens from the spelling location to the new |
1054 | // expanded location. |
1055 | for (Token& T : Partition) { |
1056 | SourceLocation::IntTy RelativeOffset = |
1057 | T.getLocation().getRawEncoding() - BeginLoc.getRawEncoding(); |
1058 | T.setLocation(Expansion.getLocWithOffset(Offset: RelativeOffset)); |
1059 | } |
1060 | begin_tokens = &Partition.back() + 1; |
1061 | } |
1062 | |
1063 | /// Creates SLocEntries and updates the locations of macro argument |
1064 | /// tokens to their new expanded locations. |
1065 | /// |
1066 | /// \param ArgIdSpellLoc the location of the macro argument id inside the macro |
1067 | /// definition. |
1068 | void TokenLexer::updateLocForMacroArgTokens(SourceLocation ArgIdSpellLoc, |
1069 | Token *begin_tokens, |
1070 | Token *end_tokens) { |
1071 | SourceManager &SM = PP.getSourceManager(); |
1072 | |
1073 | SourceLocation ExpandLoc = |
1074 | getExpansionLocForMacroDefLoc(loc: ArgIdSpellLoc); |
1075 | |
1076 | while (begin_tokens < end_tokens) { |
1077 | // If there's only one token just create a SLocEntry for it. |
1078 | if (end_tokens - begin_tokens == 1) { |
1079 | Token &Tok = *begin_tokens; |
1080 | Tok.setLocation(SM.createMacroArgExpansionLoc(SpellingLoc: Tok.getLocation(), |
1081 | ExpansionLoc: ExpandLoc, |
1082 | Length: Tok.getLength())); |
1083 | return; |
1084 | } |
1085 | |
1086 | updateConsecutiveMacroArgTokens(SM, ExpandLoc, begin_tokens, end_tokens); |
1087 | } |
1088 | } |
1089 | |
1090 | void TokenLexer::PropagateLineStartLeadingSpaceInfo(Token &Result) { |
1091 | AtStartOfLine = Result.isAtStartOfLine(); |
1092 | HasLeadingSpace = Result.hasLeadingSpace(); |
1093 | } |
1094 | |