1 | //===- DependencyDirectivesScanner.cpp ------------------------------------===// |
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 | /// \file |
10 | /// This is the interface for scanning header and source files to get the |
11 | /// minimum necessary preprocessor directives for evaluating includes. It |
12 | /// reduces the source down to #define, #include, #import, @import, and any |
13 | /// conditional preprocessor logic that contains one of those. |
14 | /// |
15 | //===----------------------------------------------------------------------===// |
16 | |
17 | #include "clang/Lex/DependencyDirectivesScanner.h" |
18 | #include "clang/Basic/CharInfo.h" |
19 | #include "clang/Basic/Diagnostic.h" |
20 | #include "clang/Lex/LexDiagnostic.h" |
21 | #include "clang/Lex/Lexer.h" |
22 | #include "clang/Lex/Pragma.h" |
23 | #include "llvm/ADT/ScopeExit.h" |
24 | #include "llvm/ADT/SmallString.h" |
25 | #include "llvm/ADT/StringMap.h" |
26 | #include "llvm/ADT/StringSwitch.h" |
27 | #include <optional> |
28 | |
29 | using namespace clang; |
30 | using namespace clang::dependency_directives_scan; |
31 | using namespace llvm; |
32 | |
33 | namespace { |
34 | |
35 | struct DirectiveWithTokens { |
36 | DirectiveKind Kind; |
37 | unsigned NumTokens; |
38 | |
39 | DirectiveWithTokens(DirectiveKind Kind, unsigned NumTokens) |
40 | : Kind(Kind), NumTokens(NumTokens) {} |
41 | }; |
42 | |
43 | /// Does an efficient "scan" of the sources to detect the presence of |
44 | /// preprocessor (or module import) directives and collects the raw lexed tokens |
45 | /// for those directives so that the \p Lexer can "replay" them when the file is |
46 | /// included. |
47 | /// |
48 | /// Note that the behavior of the raw lexer is affected by the language mode, |
49 | /// while at this point we want to do a scan and collect tokens once, |
50 | /// irrespective of the language mode that the file will get included in. To |
51 | /// compensate for that the \p Lexer, while "replaying", will adjust a token |
52 | /// where appropriate, when it could affect the preprocessor's state. |
53 | /// For example in a directive like |
54 | /// |
55 | /// \code |
56 | /// #if __has_cpp_attribute(clang::fallthrough) |
57 | /// \endcode |
58 | /// |
59 | /// The preprocessor needs to see '::' as 'tok::coloncolon' instead of 2 |
60 | /// 'tok::colon'. The \p Lexer will adjust if it sees consecutive 'tok::colon' |
61 | /// while in C++ mode. |
62 | struct Scanner { |
63 | Scanner(StringRef Input, |
64 | SmallVectorImpl<dependency_directives_scan::Token> &Tokens, |
65 | DiagnosticsEngine *Diags, SourceLocation InputSourceLoc) |
66 | : Input(Input), Tokens(Tokens), Diags(Diags), |
67 | InputSourceLoc(InputSourceLoc), LangOpts(getLangOptsForDepScanning()), |
68 | TheLexer(InputSourceLoc, LangOpts, Input.begin(), Input.begin(), |
69 | Input.end()) {} |
70 | |
71 | static LangOptions getLangOptsForDepScanning() { |
72 | LangOptions LangOpts; |
73 | // Set the lexer to use 'tok::at' for '@', instead of 'tok::unknown'. |
74 | LangOpts.ObjC = true; |
75 | LangOpts.LineComment = true; |
76 | LangOpts.RawStringLiterals = true; |
77 | // FIXME: we do not enable C11 or C++11, so we are missing u/u8/U"". |
78 | return LangOpts; |
79 | } |
80 | |
81 | /// Lex the provided source and emit the directive tokens. |
82 | /// |
83 | /// \returns True on error. |
84 | bool scan(SmallVectorImpl<Directive> &Directives); |
85 | |
86 | private: |
87 | /// Lexes next token and advances \p First and the \p Lexer. |
88 | [[nodiscard]] dependency_directives_scan::Token & |
89 | lexToken(const char *&First, const char *const End); |
90 | |
91 | [[nodiscard]] dependency_directives_scan::Token & |
92 | lexIncludeFilename(const char *&First, const char *const End); |
93 | |
94 | void skipLine(const char *&First, const char *const End); |
95 | void skipDirective(StringRef Name, const char *&First, const char *const End); |
96 | |
97 | /// Returns the spelling of a string literal or identifier after performing |
98 | /// any processing needed to handle \c clang::Token::NeedsCleaning. |
99 | StringRef cleanStringIfNeeded(const dependency_directives_scan::Token &Tok); |
100 | |
101 | /// Lexes next token and if it is identifier returns its string, otherwise |
102 | /// it skips the current line and returns \p std::nullopt. |
103 | /// |
104 | /// In any case (whatever the token kind) \p First and the \p Lexer will |
105 | /// advance beyond the token. |
106 | [[nodiscard]] std::optional<StringRef> |
107 | tryLexIdentifierOrSkipLine(const char *&First, const char *const End); |
108 | |
109 | /// Used when it is certain that next token is an identifier. |
110 | [[nodiscard]] StringRef lexIdentifier(const char *&First, |
111 | const char *const End); |
112 | |
113 | /// Lexes next token and returns true iff it is an identifier that matches \p |
114 | /// Id, otherwise it skips the current line and returns false. |
115 | /// |
116 | /// In any case (whatever the token kind) \p First and the \p Lexer will |
117 | /// advance beyond the token. |
118 | [[nodiscard]] bool isNextIdentifierOrSkipLine(StringRef Id, |
119 | const char *&First, |
120 | const char *const End); |
121 | |
122 | /// Lexes next token and returns true iff it matches the kind \p K. |
123 | /// Otherwise it skips the current line and returns false. |
124 | /// |
125 | /// In any case (whatever the token kind) \p First and the \p Lexer will |
126 | /// advance beyond the token. |
127 | [[nodiscard]] bool isNextTokenOrSkipLine(tok::TokenKind K, const char *&First, |
128 | const char *const End); |
129 | |
130 | /// Lexes next token and if it is string literal, returns its string. |
131 | /// Otherwise, it skips the current line and returns \p std::nullopt. |
132 | /// |
133 | /// In any case (whatever the token kind) \p First and the \p Lexer will |
134 | /// advance beyond the token. |
135 | [[nodiscard]] std::optional<StringRef> |
136 | tryLexStringLiteralOrSkipLine(const char *&First, const char *const End); |
137 | |
138 | [[nodiscard]] bool scanImpl(const char *First, const char *const End); |
139 | [[nodiscard]] bool lexPPLine(const char *&First, const char *const End); |
140 | [[nodiscard]] bool lexAt(const char *&First, const char *const End); |
141 | [[nodiscard]] bool lexModule(const char *&First, const char *const End); |
142 | [[nodiscard]] bool lexDefine(const char *HashLoc, const char *&First, |
143 | const char *const End); |
144 | [[nodiscard]] bool lexPragma(const char *&First, const char *const End); |
145 | [[nodiscard]] bool lex_Pragma(const char *&First, const char *const End); |
146 | [[nodiscard]] bool lexEndif(const char *&First, const char *const End); |
147 | [[nodiscard]] bool lexDefault(DirectiveKind Kind, const char *&First, |
148 | const char *const End); |
149 | [[nodiscard]] bool lexModuleDirectiveBody(DirectiveKind Kind, |
150 | const char *&First, |
151 | const char *const End); |
152 | void lexPPDirectiveBody(const char *&First, const char *const End); |
153 | |
154 | DirectiveWithTokens &pushDirective(DirectiveKind Kind) { |
155 | Tokens.append(RHS: CurDirToks); |
156 | DirsWithToks.emplace_back(Args&: Kind, Args: CurDirToks.size()); |
157 | CurDirToks.clear(); |
158 | return DirsWithToks.back(); |
159 | } |
160 | void popDirective() { |
161 | Tokens.pop_back_n(NumItems: DirsWithToks.pop_back_val().NumTokens); |
162 | } |
163 | DirectiveKind topDirective() const { |
164 | return DirsWithToks.empty() ? pp_none : DirsWithToks.back().Kind; |
165 | } |
166 | |
167 | unsigned getOffsetAt(const char *CurPtr) const { |
168 | return CurPtr - Input.data(); |
169 | } |
170 | |
171 | /// Reports a diagnostic if the diagnostic engine is provided. Always returns |
172 | /// true at the end. |
173 | bool reportError(const char *CurPtr, unsigned Err); |
174 | |
175 | StringMap<char> SplitIds; |
176 | StringRef Input; |
177 | SmallVectorImpl<dependency_directives_scan::Token> &Tokens; |
178 | DiagnosticsEngine *Diags; |
179 | SourceLocation InputSourceLoc; |
180 | |
181 | const char *LastTokenPtr = nullptr; |
182 | /// Keeps track of the tokens for the currently lexed directive. Once a |
183 | /// directive is fully lexed and "committed" then the tokens get appended to |
184 | /// \p Tokens and \p CurDirToks is cleared for the next directive. |
185 | SmallVector<dependency_directives_scan::Token, 32> CurDirToks; |
186 | /// The directives that were lexed along with the number of tokens that each |
187 | /// directive contains. The tokens of all the directives are kept in \p Tokens |
188 | /// vector, in the same order as the directives order in \p DirsWithToks. |
189 | SmallVector<DirectiveWithTokens, 64> DirsWithToks; |
190 | LangOptions LangOpts; |
191 | Lexer TheLexer; |
192 | }; |
193 | |
194 | } // end anonymous namespace |
195 | |
196 | bool Scanner::reportError(const char *CurPtr, unsigned Err) { |
197 | if (!Diags) |
198 | return true; |
199 | assert(CurPtr >= Input.data() && "invalid buffer ptr" ); |
200 | Diags->Report(Loc: InputSourceLoc.getLocWithOffset(Offset: getOffsetAt(CurPtr)), DiagID: Err); |
201 | return true; |
202 | } |
203 | |
204 | static void skipOverSpaces(const char *&First, const char *const End) { |
205 | while (First != End && isHorizontalWhitespace(c: *First)) |
206 | ++First; |
207 | } |
208 | |
209 | // Move back by one character, skipping escaped newlines (backslash + \n) |
210 | static char previousChar(const char *First, const char *&Current) { |
211 | assert(Current > First); |
212 | --Current; |
213 | while (Current > First && isVerticalWhitespace(c: *Current)) { |
214 | // Check if the previous character is a backslash |
215 | if (Current > First && *(Current - 1) == '\\') { |
216 | // Use Lexer's getEscapedNewLineSize to get the size of the escaped |
217 | // newline |
218 | unsigned EscapeSize = Lexer::getEscapedNewLineSize(P: Current); |
219 | if (EscapeSize > 0) { |
220 | // Skip back over the entire escaped newline sequence (backslash + |
221 | // newline) |
222 | Current -= (1 + EscapeSize); |
223 | } else { |
224 | break; |
225 | } |
226 | } else { |
227 | break; |
228 | } |
229 | } |
230 | return *Current; |
231 | } |
232 | |
233 | [[nodiscard]] static bool isRawStringLiteral(const char *First, |
234 | const char *Current) { |
235 | assert(First <= Current); |
236 | |
237 | // Check if we can even back up. |
238 | if (*Current != '"' || First == Current) |
239 | return false; |
240 | |
241 | // Check for an "R". |
242 | if (previousChar(First, Current) != 'R') |
243 | return false; |
244 | if (First == Current || |
245 | !isAsciiIdentifierContinue(c: previousChar(First, Current))) |
246 | return true; |
247 | |
248 | // Check for a prefix of "u", "U", or "L". |
249 | if (*Current == 'u' || *Current == 'U' || *Current == 'L') |
250 | return First == Current || |
251 | !isAsciiIdentifierContinue(c: previousChar(First, Current)); |
252 | |
253 | // Check for a prefix of "u8". |
254 | if (*Current != '8' || First == Current || |
255 | previousChar(First, Current) != 'u') |
256 | return false; |
257 | return First == Current || |
258 | !isAsciiIdentifierContinue(c: previousChar(First, Current)); |
259 | } |
260 | |
261 | static void skipRawString(const char *&First, const char *const End) { |
262 | assert(First[0] == '"'); |
263 | |
264 | const char *Last = ++First; |
265 | while (Last != End && *Last != '(') |
266 | ++Last; |
267 | if (Last == End) { |
268 | First = Last; // Hit the end... just give up. |
269 | return; |
270 | } |
271 | |
272 | StringRef Terminator(First, Last - First); |
273 | for (;;) { |
274 | // Move First to just past the next ")". |
275 | First = Last; |
276 | while (First != End && *First != ')') |
277 | ++First; |
278 | if (First == End) |
279 | return; |
280 | ++First; |
281 | |
282 | // Look ahead for the terminator sequence. |
283 | Last = First; |
284 | while (Last != End && size_t(Last - First) < Terminator.size() && |
285 | Terminator[Last - First] == *Last) |
286 | ++Last; |
287 | |
288 | // Check if we hit it (or the end of the file). |
289 | if (Last == End) { |
290 | First = Last; |
291 | return; |
292 | } |
293 | if (size_t(Last - First) < Terminator.size()) |
294 | continue; |
295 | if (*Last != '"') |
296 | continue; |
297 | First = Last + 1; |
298 | return; |
299 | } |
300 | } |
301 | |
302 | // Returns the length of EOL, either 0 (no end-of-line), 1 (\n) or 2 (\r\n) |
303 | static unsigned isEOL(const char *First, const char *const End) { |
304 | if (First == End) |
305 | return 0; |
306 | if (End - First > 1 && isVerticalWhitespace(c: First[0]) && |
307 | isVerticalWhitespace(c: First[1]) && First[0] != First[1]) |
308 | return 2; |
309 | return !!isVerticalWhitespace(c: First[0]); |
310 | } |
311 | |
312 | static void skipString(const char *&First, const char *const End) { |
313 | assert(*First == '\'' || *First == '"' || *First == '<'); |
314 | const char Terminator = *First == '<' ? '>' : *First; |
315 | for (++First; First != End && *First != Terminator; ++First) { |
316 | // String and character literals don't extend past the end of the line. |
317 | if (isVerticalWhitespace(c: *First)) |
318 | return; |
319 | if (*First != '\\') |
320 | continue; |
321 | // Skip past backslash to the next character. This ensures that the |
322 | // character right after it is skipped as well, which matters if it's |
323 | // the terminator. |
324 | if (++First == End) |
325 | return; |
326 | if (!isWhitespace(c: *First)) |
327 | continue; |
328 | // Whitespace after the backslash might indicate a line continuation. |
329 | const char *FirstAfterBackslashPastSpace = First; |
330 | skipOverSpaces(First&: FirstAfterBackslashPastSpace, End); |
331 | if (unsigned NLSize = isEOL(First: FirstAfterBackslashPastSpace, End)) { |
332 | // Advance the character pointer to the next line for the next |
333 | // iteration. |
334 | First = FirstAfterBackslashPastSpace + NLSize - 1; |
335 | } |
336 | } |
337 | if (First != End) |
338 | ++First; // Finish off the string. |
339 | } |
340 | |
341 | // Returns the length of the skipped newline |
342 | static unsigned skipNewline(const char *&First, const char *End) { |
343 | if (First == End) |
344 | return 0; |
345 | assert(isVerticalWhitespace(*First)); |
346 | unsigned Len = isEOL(First, End); |
347 | assert(Len && "expected newline" ); |
348 | First += Len; |
349 | return Len; |
350 | } |
351 | |
352 | static void skipToNewlineRaw(const char *&First, const char *const End) { |
353 | for (;;) { |
354 | if (First == End) |
355 | return; |
356 | |
357 | unsigned Len = isEOL(First, End); |
358 | if (Len) |
359 | return; |
360 | |
361 | char LastNonWhitespace = ' '; |
362 | do { |
363 | if (!isHorizontalWhitespace(c: *First)) |
364 | LastNonWhitespace = *First; |
365 | if (++First == End) |
366 | return; |
367 | Len = isEOL(First, End); |
368 | } while (!Len); |
369 | |
370 | if (LastNonWhitespace != '\\') |
371 | return; |
372 | |
373 | First += Len; |
374 | // Keep skipping lines... |
375 | } |
376 | } |
377 | |
378 | static void (const char *&First, const char *const End) { |
379 | assert(First[0] == '/' && First[1] == '/'); |
380 | First += 2; |
381 | skipToNewlineRaw(First, End); |
382 | } |
383 | |
384 | static void (const char *&First, const char *const End) { |
385 | assert(First[0] == '/' && First[1] == '*'); |
386 | if (End - First < 4) { |
387 | First = End; |
388 | return; |
389 | } |
390 | for (First += 3; First != End; ++First) |
391 | if (First[-1] == '*' && First[0] == '/') { |
392 | ++First; |
393 | return; |
394 | } |
395 | } |
396 | |
397 | /// \returns True if the current single quotation mark character is a C++14 |
398 | /// digit separator. |
399 | static bool isQuoteCppDigitSeparator(const char *const Start, |
400 | const char *const Cur, |
401 | const char *const End) { |
402 | assert(*Cur == '\'' && "expected quotation character" ); |
403 | // skipLine called in places where we don't expect a valid number |
404 | // body before `start` on the same line, so always return false at the start. |
405 | if (Start == Cur) |
406 | return false; |
407 | // The previous character must be a valid PP number character. |
408 | // Make sure that the L, u, U, u8 prefixes don't get marked as a |
409 | // separator though. |
410 | char Prev = *(Cur - 1); |
411 | if (Prev == 'L' || Prev == 'U' || Prev == 'u') |
412 | return false; |
413 | if (Prev == '8' && (Cur - 1 != Start) && *(Cur - 2) == 'u') |
414 | return false; |
415 | if (!isPreprocessingNumberBody(c: Prev)) |
416 | return false; |
417 | // The next character should be a valid identifier body character. |
418 | return (Cur + 1) < End && isAsciiIdentifierContinue(c: *(Cur + 1)); |
419 | } |
420 | |
421 | void Scanner::skipLine(const char *&First, const char *const End) { |
422 | for (;;) { |
423 | assert(First <= End); |
424 | if (First == End) |
425 | return; |
426 | |
427 | if (isVerticalWhitespace(c: *First)) { |
428 | skipNewline(First, End); |
429 | return; |
430 | } |
431 | const char *Start = First; |
432 | // Use `LastNonWhitespace`to track if a line-continuation has ever been seen |
433 | // before a new-line character: |
434 | char LastNonWhitespace = ' '; |
435 | while (First != End && !isVerticalWhitespace(c: *First)) { |
436 | // Iterate over strings correctly to avoid comments and newlines. |
437 | if (*First == '"' || |
438 | (*First == '\'' && !isQuoteCppDigitSeparator(Start, Cur: First, End))) { |
439 | LastTokenPtr = First; |
440 | if (isRawStringLiteral(First: Start, Current: First)) |
441 | skipRawString(First, End); |
442 | else |
443 | skipString(First, End); |
444 | continue; |
445 | } |
446 | |
447 | // Continue on the same line if an EOL is preceded with backslash |
448 | if (First + 1 < End && *First == '\\') { |
449 | if (unsigned Len = isEOL(First: First + 1, End)) { |
450 | First += 1 + Len; |
451 | continue; |
452 | } |
453 | } |
454 | |
455 | // Iterate over comments correctly. |
456 | if (*First != '/' || End - First < 2) { |
457 | LastTokenPtr = First; |
458 | if (!isWhitespace(c: *First)) |
459 | LastNonWhitespace = *First; |
460 | ++First; |
461 | continue; |
462 | } |
463 | |
464 | if (First[1] == '/') { |
465 | // "//...". |
466 | skipLineComment(First, End); |
467 | continue; |
468 | } |
469 | |
470 | if (First[1] != '*') { |
471 | LastTokenPtr = First; |
472 | if (!isWhitespace(c: *First)) |
473 | LastNonWhitespace = *First; |
474 | ++First; |
475 | continue; |
476 | } |
477 | |
478 | // "/*...*/". |
479 | skipBlockComment(First, End); |
480 | } |
481 | if (First == End) |
482 | return; |
483 | |
484 | // Skip over the newline. |
485 | skipNewline(First, End); |
486 | |
487 | if (LastNonWhitespace != '\\') |
488 | break; |
489 | } |
490 | } |
491 | |
492 | void Scanner::skipDirective(StringRef Name, const char *&First, |
493 | const char *const End) { |
494 | if (llvm::StringSwitch<bool>(Name) |
495 | .Case(S: "warning" , Value: true) |
496 | .Case(S: "error" , Value: true) |
497 | .Default(Value: false)) |
498 | // Do not process quotes or comments. |
499 | skipToNewlineRaw(First, End); |
500 | else |
501 | skipLine(First, End); |
502 | } |
503 | |
504 | static void skipWhitespace(const char *&First, const char *const End) { |
505 | for (;;) { |
506 | assert(First <= End); |
507 | skipOverSpaces(First, End); |
508 | |
509 | if (End - First < 2) |
510 | return; |
511 | |
512 | if (*First == '\\') { |
513 | const char *Ptr = First + 1; |
514 | while (Ptr < End && isHorizontalWhitespace(c: *Ptr)) |
515 | ++Ptr; |
516 | if (Ptr != End && isVerticalWhitespace(c: *Ptr)) { |
517 | skipNewline(First&: Ptr, End); |
518 | First = Ptr; |
519 | continue; |
520 | } |
521 | return; |
522 | } |
523 | |
524 | // Check for a non-comment character. |
525 | if (First[0] != '/') |
526 | return; |
527 | |
528 | // "// ...". |
529 | if (First[1] == '/') { |
530 | skipLineComment(First, End); |
531 | return; |
532 | } |
533 | |
534 | // Cannot be a comment. |
535 | if (First[1] != '*') |
536 | return; |
537 | |
538 | // "/*...*/". |
539 | skipBlockComment(First, End); |
540 | } |
541 | } |
542 | |
543 | bool Scanner::lexModuleDirectiveBody(DirectiveKind Kind, const char *&First, |
544 | const char *const End) { |
545 | const char *DirectiveLoc = Input.data() + CurDirToks.front().Offset; |
546 | for (;;) { |
547 | // Keep a copy of the First char incase it needs to be reset. |
548 | const char *Previous = First; |
549 | const dependency_directives_scan::Token &Tok = lexToken(First, End); |
550 | if ((Tok.is(K: tok::hash) || Tok.is(K: tok::at)) && |
551 | (Tok.Flags & clang::Token::StartOfLine)) { |
552 | CurDirToks.pop_back(); |
553 | First = Previous; |
554 | return false; |
555 | } |
556 | if (Tok.is(K: tok::eof)) |
557 | return reportError( |
558 | CurPtr: DirectiveLoc, |
559 | Err: diag::err_dep_source_scanner_missing_semi_after_at_import); |
560 | if (Tok.is(K: tok::semi)) |
561 | break; |
562 | } |
563 | pushDirective(Kind); |
564 | skipWhitespace(First, End); |
565 | if (First == End) |
566 | return false; |
567 | if (!isVerticalWhitespace(c: *First)) |
568 | return reportError( |
569 | CurPtr: DirectiveLoc, Err: diag::err_dep_source_scanner_unexpected_tokens_at_import); |
570 | skipNewline(First, End); |
571 | return false; |
572 | } |
573 | |
574 | dependency_directives_scan::Token &Scanner::lexToken(const char *&First, |
575 | const char *const End) { |
576 | clang::Token Tok; |
577 | TheLexer.LexFromRawLexer(Result&: Tok); |
578 | First = Input.data() + TheLexer.getCurrentBufferOffset(); |
579 | assert(First <= End); |
580 | |
581 | unsigned Offset = TheLexer.getCurrentBufferOffset() - Tok.getLength(); |
582 | CurDirToks.emplace_back(Args&: Offset, Args: Tok.getLength(), Args: Tok.getKind(), |
583 | Args: Tok.getFlags()); |
584 | return CurDirToks.back(); |
585 | } |
586 | |
587 | dependency_directives_scan::Token & |
588 | Scanner::lexIncludeFilename(const char *&First, const char *const End) { |
589 | clang::Token Tok; |
590 | TheLexer.LexIncludeFilename(FilenameTok&: Tok); |
591 | First = Input.data() + TheLexer.getCurrentBufferOffset(); |
592 | assert(First <= End); |
593 | |
594 | unsigned Offset = TheLexer.getCurrentBufferOffset() - Tok.getLength(); |
595 | CurDirToks.emplace_back(Args&: Offset, Args: Tok.getLength(), Args: Tok.getKind(), |
596 | Args: Tok.getFlags()); |
597 | return CurDirToks.back(); |
598 | } |
599 | |
600 | void Scanner::lexPPDirectiveBody(const char *&First, const char *const End) { |
601 | while (true) { |
602 | const dependency_directives_scan::Token &Tok = lexToken(First, End); |
603 | if (Tok.is(K: tok::eod) || Tok.is(K: tok::eof)) |
604 | break; |
605 | } |
606 | } |
607 | |
608 | StringRef |
609 | Scanner::cleanStringIfNeeded(const dependency_directives_scan::Token &Tok) { |
610 | bool NeedsCleaning = Tok.Flags & clang::Token::NeedsCleaning; |
611 | if (LLVM_LIKELY(!NeedsCleaning)) |
612 | return Input.slice(Start: Tok.Offset, End: Tok.getEnd()); |
613 | |
614 | SmallString<64> Spelling; |
615 | Spelling.resize(N: Tok.Length); |
616 | |
617 | // FIXME: C++11 raw string literals need special handling (see getSpellingSlow |
618 | // in the Lexer). Currently we cannot see them due to our LangOpts. |
619 | |
620 | unsigned SpellingLength = 0; |
621 | const char *BufPtr = Input.begin() + Tok.Offset; |
622 | const char *AfterIdent = Input.begin() + Tok.getEnd(); |
623 | while (BufPtr < AfterIdent) { |
624 | auto [Char, Size] = Lexer::getCharAndSizeNoWarn(Ptr: BufPtr, LangOpts); |
625 | Spelling[SpellingLength++] = Char; |
626 | BufPtr += Size; |
627 | } |
628 | |
629 | return SplitIds.try_emplace(Key: StringRef(Spelling.begin(), SpellingLength), Args: 0) |
630 | .first->first(); |
631 | } |
632 | |
633 | std::optional<StringRef> |
634 | Scanner::tryLexIdentifierOrSkipLine(const char *&First, const char *const End) { |
635 | const dependency_directives_scan::Token &Tok = lexToken(First, End); |
636 | if (Tok.isNot(K: tok::raw_identifier)) { |
637 | if (!Tok.is(K: tok::eod)) |
638 | skipLine(First, End); |
639 | return std::nullopt; |
640 | } |
641 | |
642 | return cleanStringIfNeeded(Tok); |
643 | } |
644 | |
645 | StringRef Scanner::lexIdentifier(const char *&First, const char *const End) { |
646 | std::optional<StringRef> Id = tryLexIdentifierOrSkipLine(First, End); |
647 | assert(Id && "expected identifier token" ); |
648 | return *Id; |
649 | } |
650 | |
651 | bool Scanner::isNextIdentifierOrSkipLine(StringRef Id, const char *&First, |
652 | const char *const End) { |
653 | if (std::optional<StringRef> FoundId = |
654 | tryLexIdentifierOrSkipLine(First, End)) { |
655 | if (*FoundId == Id) |
656 | return true; |
657 | skipLine(First, End); |
658 | } |
659 | return false; |
660 | } |
661 | |
662 | bool Scanner::isNextTokenOrSkipLine(tok::TokenKind K, const char *&First, |
663 | const char *const End) { |
664 | const dependency_directives_scan::Token &Tok = lexToken(First, End); |
665 | if (Tok.is(K)) |
666 | return true; |
667 | skipLine(First, End); |
668 | return false; |
669 | } |
670 | |
671 | std::optional<StringRef> |
672 | Scanner::tryLexStringLiteralOrSkipLine(const char *&First, |
673 | const char *const End) { |
674 | const dependency_directives_scan::Token &Tok = lexToken(First, End); |
675 | if (!tok::isStringLiteral(K: Tok.Kind)) { |
676 | if (!Tok.is(K: tok::eod)) |
677 | skipLine(First, End); |
678 | return std::nullopt; |
679 | } |
680 | |
681 | return cleanStringIfNeeded(Tok); |
682 | } |
683 | |
684 | bool Scanner::lexAt(const char *&First, const char *const End) { |
685 | // Handle "@import". |
686 | |
687 | // Lex '@'. |
688 | const dependency_directives_scan::Token &AtTok = lexToken(First, End); |
689 | assert(AtTok.is(tok::at)); |
690 | (void)AtTok; |
691 | |
692 | if (!isNextIdentifierOrSkipLine(Id: "import" , First, End)) |
693 | return false; |
694 | return lexModuleDirectiveBody(Kind: decl_at_import, First, End); |
695 | } |
696 | |
697 | bool Scanner::lexModule(const char *&First, const char *const End) { |
698 | StringRef Id = lexIdentifier(First, End); |
699 | bool Export = false; |
700 | if (Id == "export" ) { |
701 | Export = true; |
702 | std::optional<StringRef> NextId = tryLexIdentifierOrSkipLine(First, End); |
703 | if (!NextId) |
704 | return false; |
705 | Id = *NextId; |
706 | } |
707 | |
708 | if (Id != "module" && Id != "import" ) { |
709 | skipLine(First, End); |
710 | return false; |
711 | } |
712 | |
713 | skipWhitespace(First, End); |
714 | |
715 | // Ignore this as a module directive if the next character can't be part of |
716 | // an import. |
717 | |
718 | switch (*First) { |
719 | case ':': { |
720 | // `module :` is never the start of a valid module declaration. |
721 | if (Id == "module" ) { |
722 | skipLine(First, End); |
723 | return false; |
724 | } |
725 | // `import:(type)name` is a valid ObjC method decl, so check one more token. |
726 | (void)lexToken(First, End); |
727 | if (!tryLexIdentifierOrSkipLine(First, End)) |
728 | return false; |
729 | break; |
730 | } |
731 | case '<': |
732 | case '"': |
733 | break; |
734 | default: |
735 | if (!isAsciiIdentifierContinue(c: *First)) { |
736 | skipLine(First, End); |
737 | return false; |
738 | } |
739 | } |
740 | |
741 | TheLexer.seek(Offset: getOffsetAt(CurPtr: First), /*IsAtStartOfLine*/ false); |
742 | |
743 | DirectiveKind Kind; |
744 | if (Id == "module" ) |
745 | Kind = Export ? cxx_export_module_decl : cxx_module_decl; |
746 | else |
747 | Kind = Export ? cxx_export_import_decl : cxx_import_decl; |
748 | |
749 | return lexModuleDirectiveBody(Kind, First, End); |
750 | } |
751 | |
752 | bool Scanner::lex_Pragma(const char *&First, const char *const End) { |
753 | if (!isNextTokenOrSkipLine(K: tok::l_paren, First, End)) |
754 | return false; |
755 | |
756 | std::optional<StringRef> Str = tryLexStringLiteralOrSkipLine(First, End); |
757 | |
758 | if (!Str || !isNextTokenOrSkipLine(K: tok::r_paren, First, End)) |
759 | return false; |
760 | |
761 | SmallString<64> Buffer(*Str); |
762 | prepare_PragmaString(StrVal&: Buffer); |
763 | |
764 | // Use a new scanner instance since the tokens will be inside the allocated |
765 | // string. We should already have captured all the relevant tokens in the |
766 | // current scanner. |
767 | SmallVector<dependency_directives_scan::Token> DiscardTokens; |
768 | const char *Begin = Buffer.c_str(); |
769 | Scanner PragmaScanner{StringRef(Begin, Buffer.size()), DiscardTokens, Diags, |
770 | InputSourceLoc}; |
771 | |
772 | PragmaScanner.TheLexer.setParsingPreprocessorDirective(true); |
773 | if (PragmaScanner.lexPragma(First&: Begin, End: Buffer.end())) |
774 | return true; |
775 | |
776 | DirectiveKind K = PragmaScanner.topDirective(); |
777 | if (K == pp_none) { |
778 | skipLine(First, End); |
779 | return false; |
780 | } |
781 | |
782 | assert(Begin == Buffer.end()); |
783 | pushDirective(Kind: K); |
784 | return false; |
785 | } |
786 | |
787 | bool Scanner::lexPragma(const char *&First, const char *const End) { |
788 | std::optional<StringRef> FoundId = tryLexIdentifierOrSkipLine(First, End); |
789 | if (!FoundId) |
790 | return false; |
791 | |
792 | StringRef Id = *FoundId; |
793 | auto Kind = llvm::StringSwitch<DirectiveKind>(Id) |
794 | .Case(S: "once" , Value: pp_pragma_once) |
795 | .Case(S: "push_macro" , Value: pp_pragma_push_macro) |
796 | .Case(S: "pop_macro" , Value: pp_pragma_pop_macro) |
797 | .Case(S: "include_alias" , Value: pp_pragma_include_alias) |
798 | .Default(Value: pp_none); |
799 | if (Kind != pp_none) { |
800 | lexPPDirectiveBody(First, End); |
801 | pushDirective(Kind); |
802 | return false; |
803 | } |
804 | |
805 | if (Id != "clang" ) { |
806 | skipLine(First, End); |
807 | return false; |
808 | } |
809 | |
810 | FoundId = tryLexIdentifierOrSkipLine(First, End); |
811 | if (!FoundId) |
812 | return false; |
813 | Id = *FoundId; |
814 | |
815 | // #pragma clang system_header |
816 | if (Id == "system_header" ) { |
817 | lexPPDirectiveBody(First, End); |
818 | pushDirective(Kind: pp_pragma_system_header); |
819 | return false; |
820 | } |
821 | |
822 | if (Id != "module" ) { |
823 | skipLine(First, End); |
824 | return false; |
825 | } |
826 | |
827 | // #pragma clang module. |
828 | if (!isNextIdentifierOrSkipLine(Id: "import" , First, End)) |
829 | return false; |
830 | |
831 | // #pragma clang module import. |
832 | lexPPDirectiveBody(First, End); |
833 | pushDirective(Kind: pp_pragma_import); |
834 | return false; |
835 | } |
836 | |
837 | bool Scanner::lexEndif(const char *&First, const char *const End) { |
838 | // Strip out "#else" if it's empty. |
839 | if (topDirective() == pp_else) |
840 | popDirective(); |
841 | |
842 | // If "#ifdef" is empty, strip it and skip the "#endif". |
843 | // |
844 | // FIXME: Once/if Clang starts disallowing __has_include in macro expansions, |
845 | // we can skip empty `#if` and `#elif` blocks as well after scanning for a |
846 | // literal __has_include in the condition. Even without that rule we could |
847 | // drop the tokens if we scan for identifiers in the condition and find none. |
848 | if (topDirective() == pp_ifdef || topDirective() == pp_ifndef) { |
849 | popDirective(); |
850 | skipLine(First, End); |
851 | return false; |
852 | } |
853 | |
854 | return lexDefault(Kind: pp_endif, First, End); |
855 | } |
856 | |
857 | bool Scanner::lexDefault(DirectiveKind Kind, const char *&First, |
858 | const char *const End) { |
859 | lexPPDirectiveBody(First, End); |
860 | pushDirective(Kind); |
861 | return false; |
862 | } |
863 | |
864 | static bool isStartOfRelevantLine(char First) { |
865 | switch (First) { |
866 | case '#': |
867 | case '@': |
868 | case 'i': |
869 | case 'e': |
870 | case 'm': |
871 | case '_': |
872 | return true; |
873 | } |
874 | return false; |
875 | } |
876 | |
877 | bool Scanner::lexPPLine(const char *&First, const char *const End) { |
878 | assert(First != End); |
879 | |
880 | skipWhitespace(First, End); |
881 | assert(First <= End); |
882 | if (First == End) |
883 | return false; |
884 | |
885 | if (!isStartOfRelevantLine(First: *First)) { |
886 | skipLine(First, End); |
887 | assert(First <= End); |
888 | return false; |
889 | } |
890 | |
891 | LastTokenPtr = First; |
892 | |
893 | TheLexer.seek(Offset: getOffsetAt(CurPtr: First), /*IsAtStartOfLine*/ true); |
894 | |
895 | auto ScEx1 = make_scope_exit(F: [&]() { |
896 | /// Clear Scanner's CurDirToks before returning, in case we didn't push a |
897 | /// new directive. |
898 | CurDirToks.clear(); |
899 | }); |
900 | |
901 | // Handle "@import". |
902 | if (*First == '@') |
903 | return lexAt(First, End); |
904 | |
905 | // Handle module directives for C++20 modules. |
906 | if (*First == 'i' || *First == 'e' || *First == 'm') |
907 | return lexModule(First, End); |
908 | |
909 | if (*First == '_') { |
910 | if (isNextIdentifierOrSkipLine(Id: "_Pragma" , First, End)) |
911 | return lex_Pragma(First, End); |
912 | return false; |
913 | } |
914 | |
915 | // Handle preprocessing directives. |
916 | |
917 | TheLexer.setParsingPreprocessorDirective(true); |
918 | auto ScEx2 = make_scope_exit( |
919 | F: [&]() { TheLexer.setParsingPreprocessorDirective(false); }); |
920 | |
921 | // Lex '#'. |
922 | const dependency_directives_scan::Token &HashTok = lexToken(First, End); |
923 | if (HashTok.is(K: tok::hashhash)) { |
924 | // A \p tok::hashhash at this location is passed by the preprocessor to the |
925 | // parser to interpret, like any other token. So for dependency scanning |
926 | // skip it like a normal token not affecting the preprocessor. |
927 | skipLine(First, End); |
928 | assert(First <= End); |
929 | return false; |
930 | } |
931 | assert(HashTok.is(tok::hash)); |
932 | (void)HashTok; |
933 | |
934 | std::optional<StringRef> FoundId = tryLexIdentifierOrSkipLine(First, End); |
935 | if (!FoundId) |
936 | return false; |
937 | |
938 | StringRef Id = *FoundId; |
939 | |
940 | if (Id == "pragma" ) |
941 | return lexPragma(First, End); |
942 | |
943 | auto Kind = llvm::StringSwitch<DirectiveKind>(Id) |
944 | .Case(S: "include" , Value: pp_include) |
945 | .Case(S: "__include_macros" , Value: pp___include_macros) |
946 | .Case(S: "define" , Value: pp_define) |
947 | .Case(S: "undef" , Value: pp_undef) |
948 | .Case(S: "import" , Value: pp_import) |
949 | .Case(S: "include_next" , Value: pp_include_next) |
950 | .Case(S: "if" , Value: pp_if) |
951 | .Case(S: "ifdef" , Value: pp_ifdef) |
952 | .Case(S: "ifndef" , Value: pp_ifndef) |
953 | .Case(S: "elif" , Value: pp_elif) |
954 | .Case(S: "elifdef" , Value: pp_elifdef) |
955 | .Case(S: "elifndef" , Value: pp_elifndef) |
956 | .Case(S: "else" , Value: pp_else) |
957 | .Case(S: "endif" , Value: pp_endif) |
958 | .Default(Value: pp_none); |
959 | if (Kind == pp_none) { |
960 | skipDirective(Name: Id, First, End); |
961 | return false; |
962 | } |
963 | |
964 | if (Kind == pp_endif) |
965 | return lexEndif(First, End); |
966 | |
967 | switch (Kind) { |
968 | case pp_include: |
969 | case pp___include_macros: |
970 | case pp_include_next: |
971 | case pp_import: |
972 | // Ignore missing filenames in include or import directives. |
973 | if (lexIncludeFilename(First, End).is(K: tok::eod)) { |
974 | return false; |
975 | } |
976 | break; |
977 | default: |
978 | break; |
979 | } |
980 | |
981 | // Everything else. |
982 | return lexDefault(Kind, First, End); |
983 | } |
984 | |
985 | static void skipUTF8ByteOrderMark(const char *&First, const char *const End) { |
986 | if ((End - First) >= 3 && First[0] == '\xef' && First[1] == '\xbb' && |
987 | First[2] == '\xbf') |
988 | First += 3; |
989 | } |
990 | |
991 | bool Scanner::scanImpl(const char *First, const char *const End) { |
992 | skipUTF8ByteOrderMark(First, End); |
993 | while (First != End) |
994 | if (lexPPLine(First, End)) |
995 | return true; |
996 | return false; |
997 | } |
998 | |
999 | bool Scanner::scan(SmallVectorImpl<Directive> &Directives) { |
1000 | bool Error = scanImpl(First: Input.begin(), End: Input.end()); |
1001 | |
1002 | if (!Error) { |
1003 | // Add an EOF on success. |
1004 | if (LastTokenPtr && |
1005 | (Tokens.empty() || LastTokenPtr > Input.begin() + Tokens.back().Offset)) |
1006 | pushDirective(Kind: tokens_present_before_eof); |
1007 | pushDirective(Kind: pp_eof); |
1008 | } |
1009 | |
1010 | ArrayRef<dependency_directives_scan::Token> RemainingTokens = Tokens; |
1011 | for (const DirectiveWithTokens &DirWithToks : DirsWithToks) { |
1012 | assert(RemainingTokens.size() >= DirWithToks.NumTokens); |
1013 | Directives.emplace_back(Args: DirWithToks.Kind, |
1014 | Args: RemainingTokens.take_front(N: DirWithToks.NumTokens)); |
1015 | RemainingTokens = RemainingTokens.drop_front(N: DirWithToks.NumTokens); |
1016 | } |
1017 | assert(RemainingTokens.empty()); |
1018 | |
1019 | return Error; |
1020 | } |
1021 | |
1022 | bool clang::scanSourceForDependencyDirectives( |
1023 | StringRef Input, SmallVectorImpl<dependency_directives_scan::Token> &Tokens, |
1024 | SmallVectorImpl<Directive> &Directives, DiagnosticsEngine *Diags, |
1025 | SourceLocation InputSourceLoc) { |
1026 | return Scanner(Input, Tokens, Diags, InputSourceLoc).scan(Directives); |
1027 | } |
1028 | |
1029 | void clang::printDependencyDirectivesAsSource( |
1030 | StringRef Source, |
1031 | ArrayRef<dependency_directives_scan::Directive> Directives, |
1032 | llvm::raw_ostream &OS) { |
1033 | // Add a space separator where it is convenient for testing purposes. |
1034 | auto needsSpaceSeparator = |
1035 | [](tok::TokenKind Prev, |
1036 | const dependency_directives_scan::Token &Tok) -> bool { |
1037 | if (Prev == Tok.Kind) |
1038 | return !Tok.isOneOf(K1: tok::l_paren, Ks: tok::r_paren, Ks: tok::l_square, |
1039 | Ks: tok::r_square); |
1040 | if (Prev == tok::raw_identifier && |
1041 | Tok.isOneOf(K1: tok::hash, Ks: tok::numeric_constant, Ks: tok::string_literal, |
1042 | Ks: tok::char_constant, Ks: tok::header_name)) |
1043 | return true; |
1044 | if (Prev == tok::r_paren && |
1045 | Tok.isOneOf(K1: tok::raw_identifier, Ks: tok::hash, Ks: tok::string_literal, |
1046 | Ks: tok::char_constant, Ks: tok::unknown)) |
1047 | return true; |
1048 | if (Prev == tok::comma && |
1049 | Tok.isOneOf(K1: tok::l_paren, Ks: tok::string_literal, Ks: tok::less)) |
1050 | return true; |
1051 | return false; |
1052 | }; |
1053 | |
1054 | for (const dependency_directives_scan::Directive &Directive : Directives) { |
1055 | if (Directive.Kind == tokens_present_before_eof) |
1056 | OS << "<TokBeforeEOF>" ; |
1057 | std::optional<tok::TokenKind> PrevTokenKind; |
1058 | for (const dependency_directives_scan::Token &Tok : Directive.Tokens) { |
1059 | if (PrevTokenKind && needsSpaceSeparator(*PrevTokenKind, Tok)) |
1060 | OS << ' '; |
1061 | PrevTokenKind = Tok.Kind; |
1062 | OS << Source.slice(Start: Tok.Offset, End: Tok.getEnd()); |
1063 | } |
1064 | } |
1065 | } |
1066 | |