1//===--- UnwrappedLineParser.cpp - Format C++ code ------------------------===//
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 file contains the implementation of the UnwrappedLineParser,
11/// which turns a stream of tokens into UnwrappedLines.
12///
13//===----------------------------------------------------------------------===//
14
15#include "UnwrappedLineParser.h"
16#include "FormatToken.h"
17#include "FormatTokenSource.h"
18#include "Macros.h"
19#include "TokenAnnotator.h"
20#include "clang/Basic/TokenKinds.h"
21#include "llvm/ADT/STLExtras.h"
22#include "llvm/ADT/StringRef.h"
23#include "llvm/Support/Debug.h"
24#include "llvm/Support/raw_os_ostream.h"
25#include "llvm/Support/raw_ostream.h"
26
27#include <utility>
28
29#define DEBUG_TYPE "format-parser"
30
31namespace clang {
32namespace format {
33
34namespace {
35
36void printLine(llvm::raw_ostream &OS, const UnwrappedLine &Line,
37 StringRef Prefix = "", bool PrintText = false) {
38 OS << Prefix << "Line(" << Line.Level << ", FSC=" << Line.FirstStartColumn
39 << ")" << (Line.InPPDirective ? " MACRO" : "") << ": ";
40 bool NewLine = false;
41 for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(),
42 E = Line.Tokens.end();
43 I != E; ++I) {
44 if (NewLine) {
45 OS << Prefix;
46 NewLine = false;
47 }
48 OS << I->Tok->Tok.getName() << "["
49 << "T=" << (unsigned)I->Tok->getType()
50 << ", OC=" << I->Tok->OriginalColumn << ", \"" << I->Tok->TokenText
51 << "\"] ";
52 for (const auto *CI = I->Children.begin(), *CE = I->Children.end();
53 CI != CE; ++CI) {
54 OS << "\n";
55 printLine(OS, Line: *CI, Prefix: (Prefix + " ").str());
56 NewLine = true;
57 }
58 }
59 if (!NewLine)
60 OS << "\n";
61}
62
63[[maybe_unused]] static void printDebugInfo(const UnwrappedLine &Line) {
64 printLine(OS&: llvm::dbgs(), Line);
65}
66
67class ScopedDeclarationState {
68public:
69 ScopedDeclarationState(UnwrappedLine &Line, llvm::BitVector &Stack,
70 bool MustBeDeclaration)
71 : Line(Line), Stack(Stack) {
72 Line.MustBeDeclaration = MustBeDeclaration;
73 Stack.push_back(Val: MustBeDeclaration);
74 }
75 ~ScopedDeclarationState() {
76 Stack.pop_back();
77 if (!Stack.empty())
78 Line.MustBeDeclaration = Stack.back();
79 else
80 Line.MustBeDeclaration = true;
81 }
82
83private:
84 UnwrappedLine &Line;
85 llvm::BitVector &Stack;
86};
87
88} // end anonymous namespace
89
90std::ostream &operator<<(std::ostream &Stream, const UnwrappedLine &Line) {
91 llvm::raw_os_ostream OS(Stream);
92 printLine(OS, Line);
93 return Stream;
94}
95
96class ScopedLineState {
97public:
98 ScopedLineState(UnwrappedLineParser &Parser,
99 bool SwitchToPreprocessorLines = false)
100 : Parser(Parser), OriginalLines(Parser.CurrentLines) {
101 if (SwitchToPreprocessorLines)
102 Parser.CurrentLines = &Parser.PreprocessorDirectives;
103 else if (!Parser.Line->Tokens.empty())
104 Parser.CurrentLines = &Parser.Line->Tokens.back().Children;
105 PreBlockLine = std::move(Parser.Line);
106 Parser.Line = std::make_unique<UnwrappedLine>();
107 Parser.Line->Level = PreBlockLine->Level;
108 Parser.Line->PPLevel = PreBlockLine->PPLevel;
109 Parser.Line->InPPDirective = PreBlockLine->InPPDirective;
110 Parser.Line->InMacroBody = PreBlockLine->InMacroBody;
111 Parser.Line->UnbracedBodyLevel = PreBlockLine->UnbracedBodyLevel;
112 }
113
114 ~ScopedLineState() {
115 if (!Parser.Line->Tokens.empty())
116 Parser.addUnwrappedLine();
117 assert(Parser.Line->Tokens.empty());
118 Parser.Line = std::move(PreBlockLine);
119 if (Parser.CurrentLines == &Parser.PreprocessorDirectives)
120 Parser.AtEndOfPPLine = true;
121 Parser.CurrentLines = OriginalLines;
122 }
123
124private:
125 UnwrappedLineParser &Parser;
126
127 std::unique_ptr<UnwrappedLine> PreBlockLine;
128 SmallVectorImpl<UnwrappedLine> *OriginalLines;
129};
130
131class CompoundStatementIndenter {
132public:
133 CompoundStatementIndenter(UnwrappedLineParser *Parser,
134 const FormatStyle &Style, unsigned &LineLevel)
135 : CompoundStatementIndenter(Parser, LineLevel,
136 Style.BraceWrapping.AfterControlStatement ==
137 FormatStyle::BWACS_Always,
138 Style.BraceWrapping.IndentBraces) {}
139 CompoundStatementIndenter(UnwrappedLineParser *Parser, unsigned &LineLevel,
140 bool WrapBrace, bool IndentBrace)
141 : LineLevel(LineLevel), OldLineLevel(LineLevel) {
142 if (WrapBrace)
143 Parser->addUnwrappedLine();
144 if (IndentBrace)
145 ++LineLevel;
146 }
147 ~CompoundStatementIndenter() { LineLevel = OldLineLevel; }
148
149private:
150 unsigned &LineLevel;
151 unsigned OldLineLevel;
152};
153
154UnwrappedLineParser::UnwrappedLineParser(
155 SourceManager &SourceMgr, const FormatStyle &Style,
156 const AdditionalKeywords &Keywords, unsigned FirstStartColumn,
157 ArrayRef<FormatToken *> Tokens, UnwrappedLineConsumer &Callback,
158 llvm::SpecificBumpPtrAllocator<FormatToken> &Allocator,
159 IdentifierTable &IdentTable)
160 : Line(new UnwrappedLine), AtEndOfPPLine(false), CurrentLines(&Lines),
161 Style(Style), IsCpp(Style.isCpp()),
162 LangOpts(getFormattingLangOpts(Style)), Keywords(Keywords),
163 CommentPragmasRegex(Style.CommentPragmas), Tokens(nullptr),
164 Callback(Callback), AllTokens(Tokens), PPBranchLevel(-1),
165 IncludeGuard(getIncludeGuardState(Style: Style.IndentPPDirectives)),
166 IncludeGuardToken(nullptr), FirstStartColumn(FirstStartColumn),
167 Macros(Style.Macros, SourceMgr, Style, Allocator, IdentTable) {}
168
169void UnwrappedLineParser::reset() {
170 PPBranchLevel = -1;
171 IncludeGuard = getIncludeGuardState(Style: Style.IndentPPDirectives);
172 IncludeGuardToken = nullptr;
173 Line.reset(p: new UnwrappedLine);
174 CommentsBeforeNextToken.clear();
175 FormatTok = nullptr;
176 AtEndOfPPLine = false;
177 IsDecltypeAutoFunction = false;
178 PreprocessorDirectives.clear();
179 CurrentLines = &Lines;
180 DeclarationScopeStack.clear();
181 NestedTooDeep.clear();
182 NestedLambdas.clear();
183 PPStack.clear();
184 Line->FirstStartColumn = FirstStartColumn;
185
186 if (!Unexpanded.empty())
187 for (FormatToken *Token : AllTokens)
188 Token->MacroCtx.reset();
189 CurrentExpandedLines.clear();
190 ExpandedLines.clear();
191 Unexpanded.clear();
192 InExpansion = false;
193 Reconstruct.reset();
194}
195
196void UnwrappedLineParser::parse() {
197 IndexedTokenSource TokenSource(AllTokens);
198 Line->FirstStartColumn = FirstStartColumn;
199 do {
200 LLVM_DEBUG(llvm::dbgs() << "----\n");
201 reset();
202 Tokens = &TokenSource;
203 TokenSource.reset();
204
205 readToken();
206 parseFile();
207
208 // If we found an include guard then all preprocessor directives (other than
209 // the guard) are over-indented by one.
210 if (IncludeGuard == IG_Found) {
211 for (auto &Line : Lines)
212 if (Line.InPPDirective && Line.Level > 0)
213 --Line.Level;
214 }
215
216 // Create line with eof token.
217 assert(eof());
218 pushToken(Tok: FormatTok);
219 addUnwrappedLine();
220
221 // In a first run, format everything with the lines containing macro calls
222 // replaced by the expansion.
223 if (!ExpandedLines.empty()) {
224 LLVM_DEBUG(llvm::dbgs() << "Expanded lines:\n");
225 for (const auto &Line : Lines) {
226 if (!Line.Tokens.empty()) {
227 auto it = ExpandedLines.find(Val: Line.Tokens.begin()->Tok);
228 if (it != ExpandedLines.end()) {
229 for (const auto &Expanded : it->second) {
230 LLVM_DEBUG(printDebugInfo(Expanded));
231 Callback.consumeUnwrappedLine(Line: Expanded);
232 }
233 continue;
234 }
235 }
236 LLVM_DEBUG(printDebugInfo(Line));
237 Callback.consumeUnwrappedLine(Line);
238 }
239 Callback.finishRun();
240 }
241
242 LLVM_DEBUG(llvm::dbgs() << "Unwrapped lines:\n");
243 for (const UnwrappedLine &Line : Lines) {
244 LLVM_DEBUG(printDebugInfo(Line));
245 Callback.consumeUnwrappedLine(Line);
246 }
247 Callback.finishRun();
248 Lines.clear();
249 while (!PPLevelBranchIndex.empty() &&
250 PPLevelBranchIndex.back() + 1 >= PPLevelBranchCount.back()) {
251 PPLevelBranchIndex.resize(N: PPLevelBranchIndex.size() - 1);
252 PPLevelBranchCount.resize(N: PPLevelBranchCount.size() - 1);
253 }
254 if (!PPLevelBranchIndex.empty()) {
255 ++PPLevelBranchIndex.back();
256 assert(PPLevelBranchIndex.size() == PPLevelBranchCount.size());
257 assert(PPLevelBranchIndex.back() <= PPLevelBranchCount.back());
258 }
259 } while (!PPLevelBranchIndex.empty());
260}
261
262void UnwrappedLineParser::parseFile() {
263 // The top-level context in a file always has declarations, except for pre-
264 // processor directives and JavaScript files.
265 bool MustBeDeclaration = !Line->InPPDirective && !Style.isJavaScript();
266 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack,
267 MustBeDeclaration);
268 if (Style.isTextProto() || (Style.isJson() && FormatTok->IsFirst))
269 parseBracedList();
270 else
271 parseLevel();
272 // Make sure to format the remaining tokens.
273 //
274 // LK_TextProto is special since its top-level is parsed as the body of a
275 // braced list, which does not necessarily have natural line separators such
276 // as a semicolon. Comments after the last entry that have been determined to
277 // not belong to that line, as in:
278 // key: value
279 // // endfile comment
280 // do not have a chance to be put on a line of their own until this point.
281 // Here we add this newline before end-of-file comments.
282 if (Style.isTextProto() && !CommentsBeforeNextToken.empty())
283 addUnwrappedLine();
284 flushComments(NewlineBeforeNext: true);
285 addUnwrappedLine();
286}
287
288void UnwrappedLineParser::parseCSharpGenericTypeConstraint() {
289 do {
290 switch (FormatTok->Tok.getKind()) {
291 case tok::l_brace:
292 case tok::semi:
293 return;
294 default:
295 if (FormatTok->is(II: Keywords.kw_where)) {
296 addUnwrappedLine();
297 nextToken();
298 parseCSharpGenericTypeConstraint();
299 break;
300 }
301 nextToken();
302 break;
303 }
304 } while (!eof());
305}
306
307void UnwrappedLineParser::parseCSharpAttribute() {
308 int UnpairedSquareBrackets = 1;
309 do {
310 switch (FormatTok->Tok.getKind()) {
311 case tok::r_square:
312 nextToken();
313 --UnpairedSquareBrackets;
314 if (UnpairedSquareBrackets == 0) {
315 addUnwrappedLine();
316 return;
317 }
318 break;
319 case tok::l_square:
320 ++UnpairedSquareBrackets;
321 nextToken();
322 break;
323 default:
324 nextToken();
325 break;
326 }
327 } while (!eof());
328}
329
330bool UnwrappedLineParser::precededByCommentOrPPDirective() const {
331 if (!Lines.empty() && Lines.back().InPPDirective)
332 return true;
333
334 const FormatToken *Previous = Tokens->getPreviousToken();
335 return Previous && Previous->is(Kind: tok::comment) &&
336 (Previous->IsMultiline || Previous->NewlinesBefore > 0);
337}
338
339/// Parses a level, that is ???.
340/// \param OpeningBrace Opening brace (\p nullptr if absent) of that level.
341/// \param IfKind The \p if statement kind in the level.
342/// \param IfLeftBrace The left brace of the \p if block in the level.
343/// \returns true if a simple block of if/else/for/while, or false otherwise.
344/// (A simple block has a single statement.)
345bool UnwrappedLineParser::parseLevel(const FormatToken *OpeningBrace,
346 IfStmtKind *IfKind,
347 FormatToken **IfLeftBrace) {
348 const bool InRequiresExpression =
349 OpeningBrace && OpeningBrace->is(TT: TT_RequiresExpressionLBrace);
350 const bool IsPrecededByCommentOrPPDirective =
351 !Style.RemoveBracesLLVM || precededByCommentOrPPDirective();
352 FormatToken *IfLBrace = nullptr;
353 bool HasDoWhile = false;
354 bool HasLabel = false;
355 unsigned StatementCount = 0;
356 bool SwitchLabelEncountered = false;
357
358 do {
359 if (FormatTok->isAttribute()) {
360 nextToken();
361 if (FormatTok->is(Kind: tok::l_paren))
362 parseParens();
363 continue;
364 }
365 tok::TokenKind Kind = FormatTok->Tok.getKind();
366 if (FormatTok->is(TT: TT_MacroBlockBegin))
367 Kind = tok::l_brace;
368 else if (FormatTok->is(TT: TT_MacroBlockEnd))
369 Kind = tok::r_brace;
370
371 auto ParseDefault = [this, OpeningBrace, IfKind, &IfLBrace, &HasDoWhile,
372 &HasLabel, &StatementCount] {
373 parseStructuralElement(OpeningBrace, IfKind, IfLeftBrace: &IfLBrace,
374 HasDoWhile: HasDoWhile ? nullptr : &HasDoWhile,
375 HasLabel: HasLabel ? nullptr : &HasLabel);
376 ++StatementCount;
377 assert(StatementCount > 0 && "StatementCount overflow!");
378 };
379
380 switch (Kind) {
381 case tok::comment:
382 nextToken();
383 addUnwrappedLine();
384 break;
385 case tok::l_brace:
386 if (InRequiresExpression) {
387 FormatTok->setFinalizedType(TT_CompoundRequirementLBrace);
388 } else if (FormatTok->Previous &&
389 FormatTok->Previous->ClosesRequiresClause) {
390 // We need the 'default' case here to correctly parse a function
391 // l_brace.
392 ParseDefault();
393 continue;
394 }
395 if (!InRequiresExpression && FormatTok->isNot(Kind: TT_MacroBlockBegin)) {
396 if (tryToParseBracedList())
397 continue;
398 FormatTok->setFinalizedType(TT_BlockLBrace);
399 }
400 parseBlock();
401 ++StatementCount;
402 assert(StatementCount > 0 && "StatementCount overflow!");
403 addUnwrappedLine();
404 break;
405 case tok::r_brace:
406 if (OpeningBrace) {
407 if (!Style.RemoveBracesLLVM || Line->InPPDirective ||
408 OpeningBrace->isNoneOf(Ks: TT_ControlStatementLBrace, Ks: TT_ElseLBrace)) {
409 return false;
410 }
411 if (FormatTok->isNot(Kind: tok::r_brace) || StatementCount != 1 || HasLabel ||
412 HasDoWhile || IsPrecededByCommentOrPPDirective ||
413 precededByCommentOrPPDirective()) {
414 return false;
415 }
416 const FormatToken *Next = Tokens->peekNextToken();
417 if (Next->is(Kind: tok::comment) && Next->NewlinesBefore == 0)
418 return false;
419 if (IfLeftBrace)
420 *IfLeftBrace = IfLBrace;
421 return true;
422 }
423 nextToken();
424 addUnwrappedLine();
425 break;
426 case tok::kw_default: {
427 unsigned StoredPosition = Tokens->getPosition();
428 auto *Next = Tokens->getNextNonComment();
429 FormatTok = Tokens->setPosition(StoredPosition);
430 if (Next->isNoneOf(Ks: tok::colon, Ks: tok::arrow)) {
431 // default not followed by `:` or `->` is not a case label; treat it
432 // like an identifier.
433 parseStructuralElement();
434 break;
435 }
436 // Else, if it is 'default:', fall through to the case handling.
437 [[fallthrough]];
438 }
439 case tok::kw_case:
440 if (Style.Language == FormatStyle::LK_Proto || Style.isVerilog() ||
441 (Style.isJavaScript() && Line->MustBeDeclaration)) {
442 // Proto: there are no switch/case statements
443 // Verilog: Case labels don't have this word. We handle case
444 // labels including default in TokenAnnotator.
445 // JavaScript: A 'case: string' style field declaration.
446 ParseDefault();
447 break;
448 }
449 if (!SwitchLabelEncountered &&
450 (Style.IndentCaseLabels ||
451 (OpeningBrace && OpeningBrace->is(TT: TT_SwitchExpressionLBrace)) ||
452 (Line->InPPDirective && Line->Level == 1))) {
453 ++Line->Level;
454 }
455 SwitchLabelEncountered = true;
456 parseStructuralElement();
457 break;
458 case tok::l_square:
459 if (Style.isCSharp()) {
460 nextToken();
461 parseCSharpAttribute();
462 break;
463 }
464 if (handleCppAttributes())
465 break;
466 [[fallthrough]];
467 default:
468 ParseDefault();
469 break;
470 }
471 } while (!eof());
472
473 return false;
474}
475
476void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) {
477 // We'll parse forward through the tokens until we hit
478 // a closing brace or eof - note that getNextToken() will
479 // parse macros, so this will magically work inside macro
480 // definitions, too.
481 unsigned StoredPosition = Tokens->getPosition();
482 FormatToken *Tok = FormatTok;
483 const FormatToken *PrevTok = Tok->Previous;
484 // Keep a stack of positions of lbrace tokens. We will
485 // update information about whether an lbrace starts a
486 // braced init list or a different block during the loop.
487 struct StackEntry {
488 FormatToken *Tok;
489 const FormatToken *PrevTok;
490 };
491 SmallVector<StackEntry, 8> LBraceStack;
492 assert(Tok->is(tok::l_brace));
493
494 do {
495 auto *NextTok = Tokens->getNextNonComment();
496
497 if (!Line->InMacroBody && !Style.isTableGen()) {
498 // Skip PPDirective lines (except macro definitions) and comments.
499 while (NextTok->is(Kind: tok::hash)) {
500 NextTok = Tokens->getNextToken();
501 if (NextTok->isOneOf(K1: tok::pp_not_keyword, K2: tok::pp_define))
502 break;
503 do {
504 NextTok = Tokens->getNextToken();
505 } while (!NextTok->HasUnescapedNewline && NextTok->isNot(Kind: tok::eof));
506
507 while (NextTok->is(Kind: tok::comment))
508 NextTok = Tokens->getNextToken();
509 }
510 }
511
512 switch (Tok->Tok.getKind()) {
513 case tok::l_brace:
514 if (Style.isJavaScript() && PrevTok) {
515 if (PrevTok->isOneOf(K1: tok::colon, K2: tok::less)) {
516 // A ':' indicates this code is in a type, or a braced list
517 // following a label in an object literal ({a: {b: 1}}).
518 // A '<' could be an object used in a comparison, but that is nonsense
519 // code (can never return true), so more likely it is a generic type
520 // argument (`X<{a: string; b: number}>`).
521 // The code below could be confused by semicolons between the
522 // individual members in a type member list, which would normally
523 // trigger BK_Block. In both cases, this must be parsed as an inline
524 // braced init.
525 Tok->setBlockKind(BK_BracedInit);
526 } else if (PrevTok->is(Kind: tok::r_paren)) {
527 // `) { }` can only occur in function or method declarations in JS.
528 Tok->setBlockKind(BK_Block);
529 }
530 } else if (Style.isJava() && PrevTok && PrevTok->is(Kind: tok::arrow)) {
531 Tok->setBlockKind(BK_Block);
532 } else {
533 Tok->setBlockKind(BK_Unknown);
534 }
535 LBraceStack.push_back(Elt: {.Tok: Tok, .PrevTok: PrevTok});
536 break;
537 case tok::r_brace:
538 if (LBraceStack.empty())
539 break;
540 if (auto *LBrace = LBraceStack.back().Tok; LBrace->is(BBK: BK_Unknown)) {
541 bool ProbablyBracedList = false;
542 if (Style.Language == FormatStyle::LK_Proto) {
543 ProbablyBracedList = NextTok->isOneOf(K1: tok::comma, K2: tok::r_square);
544 } else if (LBrace->isNot(Kind: TT_EnumLBrace)) {
545 // Using OriginalColumn to distinguish between ObjC methods and
546 // binary operators is a bit hacky.
547 bool NextIsObjCMethod = NextTok->isOneOf(K1: tok::plus, K2: tok::minus) &&
548 NextTok->OriginalColumn == 0;
549
550 // Try to detect a braced list. Note that regardless how we mark inner
551 // braces here, we will overwrite the BlockKind later if we parse a
552 // braced list (where all blocks inside are by default braced lists),
553 // or when we explicitly detect blocks (for example while parsing
554 // lambdas).
555
556 // If we already marked the opening brace as braced list, the closing
557 // must also be part of it.
558 ProbablyBracedList = LBrace->is(TT: TT_BracedListLBrace);
559
560 ProbablyBracedList = ProbablyBracedList ||
561 (Style.isJavaScript() &&
562 NextTok->isOneOf(K1: Keywords.kw_of, K2: Keywords.kw_in,
563 Ks: Keywords.kw_as));
564 ProbablyBracedList =
565 ProbablyBracedList ||
566 (IsCpp && (PrevTok->Tok.isLiteral() ||
567 NextTok->isOneOf(K1: tok::l_paren, K2: tok::arrow)));
568
569 // If there is a comma, or right paren after the closing brace, we
570 // assume this is a braced initializer list.
571 // FIXME: Some of these do not apply to JS, e.g. "} {" can never be a
572 // braced list in JS.
573 ProbablyBracedList =
574 ProbablyBracedList ||
575 NextTok->isOneOf(K1: tok::comma, K2: tok::period, Ks: tok::colon,
576 Ks: tok::r_paren, Ks: tok::r_square, Ks: tok::ellipsis);
577
578 // Distinguish between braced list in a constructor initializer list
579 // followed by constructor body, or just adjacent blocks.
580 ProbablyBracedList =
581 ProbablyBracedList ||
582 (NextTok->is(Kind: tok::l_brace) && LBraceStack.back().PrevTok &&
583 LBraceStack.back().PrevTok->isOneOf(K1: tok::identifier,
584 K2: tok::greater));
585
586 ProbablyBracedList =
587 ProbablyBracedList ||
588 (NextTok->is(Kind: tok::identifier) &&
589 PrevTok->isNoneOf(Ks: tok::semi, Ks: tok::r_brace, Ks: tok::l_brace));
590
591 ProbablyBracedList = ProbablyBracedList ||
592 (NextTok->is(Kind: tok::semi) &&
593 (!ExpectClassBody || LBraceStack.size() != 1));
594
595 ProbablyBracedList =
596 ProbablyBracedList ||
597 (NextTok->isBinaryOperator() && !NextIsObjCMethod);
598
599 if (!Style.isCSharp() && NextTok->is(Kind: tok::l_square)) {
600 // We can have an array subscript after a braced init
601 // list, but C++11 attributes are expected after blocks.
602 NextTok = Tokens->getNextToken();
603 ProbablyBracedList = NextTok->isNot(Kind: tok::l_square);
604 }
605
606 // Cpp macro definition body that is a nonempty braced list or block:
607 if (IsCpp && Line->InMacroBody && PrevTok != FormatTok &&
608 !FormatTok->Previous && NextTok->is(Kind: tok::eof) &&
609 // A statement can end with only `;` (simple statement), a block
610 // closing brace (compound statement), or `:` (label statement).
611 // If PrevTok is a block opening brace, Tok ends an empty block.
612 PrevTok->isNoneOf(Ks: tok::semi, Ks: BK_Block, Ks: tok::colon)) {
613 ProbablyBracedList = true;
614 }
615 }
616 const auto BlockKind = ProbablyBracedList ? BK_BracedInit : BK_Block;
617 Tok->setBlockKind(BlockKind);
618 LBrace->setBlockKind(BlockKind);
619 }
620 LBraceStack.pop_back();
621 break;
622 case tok::identifier:
623 if (Tok->isNot(Kind: TT_StatementMacro))
624 break;
625 [[fallthrough]];
626 case tok::at:
627 case tok::semi:
628 case tok::kw_if:
629 case tok::kw_while:
630 case tok::kw_for:
631 case tok::kw_switch:
632 case tok::kw_try:
633 case tok::kw___try:
634 if (!LBraceStack.empty() && LBraceStack.back().Tok->is(BBK: BK_Unknown))
635 LBraceStack.back().Tok->setBlockKind(BK_Block);
636 break;
637 default:
638 break;
639 }
640
641 PrevTok = Tok;
642 Tok = NextTok;
643 } while (Tok->isNot(Kind: tok::eof) && !LBraceStack.empty());
644
645 // Assume other blocks for all unclosed opening braces.
646 for (const auto &Entry : LBraceStack)
647 if (Entry.Tok->is(BBK: BK_Unknown))
648 Entry.Tok->setBlockKind(BK_Block);
649
650 FormatTok = Tokens->setPosition(StoredPosition);
651}
652
653// Sets the token type of the directly previous right brace.
654void UnwrappedLineParser::setPreviousRBraceType(TokenType Type) {
655 if (auto Prev = FormatTok->getPreviousNonComment();
656 Prev && Prev->is(Kind: tok::r_brace)) {
657 Prev->setFinalizedType(Type);
658 }
659}
660
661template <class T>
662static inline void hash_combine(std::size_t &seed, const T &v) {
663 std::hash<T> hasher;
664 seed ^= hasher(v) + 0x9e3779b9 + (seed << 6) + (seed >> 2);
665}
666
667size_t UnwrappedLineParser::computePPHash() const {
668 size_t h = 0;
669 for (const auto &i : PPStack) {
670 hash_combine(seed&: h, v: size_t(i.Kind));
671 hash_combine(seed&: h, v: i.Line);
672 }
673 return h;
674}
675
676// Checks whether \p ParsedLine might fit on a single line. If \p OpeningBrace
677// is not null, subtracts its length (plus the preceding space) when computing
678// the length of \p ParsedLine. We must clone the tokens of \p ParsedLine before
679// running the token annotator on it so that we can restore them afterward.
680bool UnwrappedLineParser::mightFitOnOneLine(
681 UnwrappedLine &ParsedLine, const FormatToken *OpeningBrace) const {
682 const auto ColumnLimit = Style.ColumnLimit;
683 if (ColumnLimit == 0)
684 return true;
685
686 auto &Tokens = ParsedLine.Tokens;
687 assert(!Tokens.empty());
688
689 const auto *LastToken = Tokens.back().Tok;
690 assert(LastToken);
691
692 SmallVector<UnwrappedLineNode> SavedTokens(Tokens.size());
693
694 int Index = 0;
695 for (const auto &Token : Tokens) {
696 assert(Token.Tok);
697 auto &SavedToken = SavedTokens[Index++];
698 SavedToken.Tok = new FormatToken;
699 SavedToken.Tok->copyFrom(Tok: *Token.Tok);
700 SavedToken.Children = std::move(Token.Children);
701 }
702
703 AnnotatedLine Line(ParsedLine);
704 assert(Line.Last == LastToken);
705
706 TokenAnnotator Annotator(Style, Keywords);
707 Annotator.annotate(Line);
708 Annotator.calculateFormattingInformation(Line);
709
710 auto Length = LastToken->TotalLength;
711 if (OpeningBrace) {
712 assert(OpeningBrace != Tokens.front().Tok);
713 if (auto Prev = OpeningBrace->Previous;
714 Prev && Prev->TotalLength + ColumnLimit == OpeningBrace->TotalLength) {
715 Length -= ColumnLimit;
716 }
717 Length -= OpeningBrace->TokenText.size() + 1;
718 }
719
720 if (const auto *FirstToken = Line.First; FirstToken->is(Kind: tok::r_brace)) {
721 assert(!OpeningBrace || OpeningBrace->is(TT_ControlStatementLBrace));
722 Length -= FirstToken->TokenText.size() + 1;
723 }
724
725 Index = 0;
726 for (auto &Token : Tokens) {
727 const auto &SavedToken = SavedTokens[Index++];
728 Token.Tok->copyFrom(Tok: *SavedToken.Tok);
729 Token.Children = std::move(SavedToken.Children);
730 delete SavedToken.Tok;
731 }
732
733 // If these change PPLevel needs to be used for get correct indentation.
734 assert(!Line.InMacroBody);
735 assert(!Line.InPPDirective);
736 return Line.Level * Style.IndentWidth + Length <= ColumnLimit;
737}
738
739FormatToken *UnwrappedLineParser::parseBlock(bool MustBeDeclaration,
740 unsigned AddLevels, bool MunchSemi,
741 bool KeepBraces,
742 IfStmtKind *IfKind,
743 bool UnindentWhitesmithsBraces) {
744 auto HandleVerilogBlockLabel = [this]() {
745 // ":" name
746 if (Style.isVerilog() && FormatTok->is(Kind: tok::colon)) {
747 nextToken();
748 if (Keywords.isVerilogIdentifier(Tok: *FormatTok))
749 nextToken();
750 }
751 };
752
753 // Whether this is a Verilog-specific block that has a special header like a
754 // module.
755 const bool VerilogHierarchy =
756 Style.isVerilog() && Keywords.isVerilogHierarchy(Tok: *FormatTok);
757 assert((FormatTok->isOneOf(tok::l_brace, TT_MacroBlockBegin) ||
758 (Style.isVerilog() &&
759 (Keywords.isVerilogBegin(*FormatTok) || VerilogHierarchy))) &&
760 "'{' or macro block token expected");
761 FormatToken *Tok = FormatTok;
762 const bool FollowedByComment = Tokens->peekNextToken()->is(Kind: tok::comment);
763 auto Index = CurrentLines->size();
764 const bool MacroBlock = FormatTok->is(TT: TT_MacroBlockBegin);
765 FormatTok->setBlockKind(BK_Block);
766
767 const bool IsWhitesmiths =
768 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths;
769
770 // For Whitesmiths mode, jump to the next level prior to skipping over the
771 // braces.
772 if (!VerilogHierarchy && AddLevels > 0 && IsWhitesmiths)
773 ++Line->Level;
774
775 size_t PPStartHash = computePPHash();
776
777 const unsigned InitialLevel = Line->Level;
778 if (VerilogHierarchy) {
779 AddLevels += parseVerilogHierarchyHeader();
780 } else {
781 nextToken(/*LevelDifference=*/AddLevels);
782 HandleVerilogBlockLabel();
783 }
784
785 // Bail out if there are too many levels. Otherwise, the stack might overflow.
786 if (Line->Level > 300)
787 return nullptr;
788
789 if (MacroBlock && FormatTok->is(Kind: tok::l_paren))
790 parseParens();
791
792 size_t NbPreprocessorDirectives =
793 !parsingPPDirective() ? PreprocessorDirectives.size() : 0;
794 addUnwrappedLine();
795 size_t OpeningLineIndex =
796 CurrentLines->empty()
797 ? (UnwrappedLine::kInvalidIndex)
798 : (CurrentLines->size() - 1 - NbPreprocessorDirectives);
799
800 // Whitesmiths is weird here. The brace needs to be indented for the namespace
801 // block, but the block itself may not be indented depending on the style
802 // settings. This allows the format to back up one level in those cases.
803 if (UnindentWhitesmithsBraces)
804 --Line->Level;
805
806 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack,
807 MustBeDeclaration);
808
809 // Whitesmiths logic has already added a level by this point, so avoid
810 // adding it twice.
811 if (AddLevels > 0u)
812 Line->Level += AddLevels - (IsWhitesmiths ? 1 : 0);
813
814 FormatToken *IfLBrace = nullptr;
815 const bool SimpleBlock = parseLevel(OpeningBrace: Tok, IfKind, IfLeftBrace: &IfLBrace);
816
817 if (eof())
818 return IfLBrace;
819
820 if (MacroBlock ? FormatTok->isNot(Kind: TT_MacroBlockEnd)
821 : FormatTok->isNot(Kind: tok::r_brace)) {
822 Line->Level = InitialLevel;
823 FormatTok->setBlockKind(BK_Block);
824 return IfLBrace;
825 }
826
827 if (FormatTok->is(Kind: tok::r_brace)) {
828 FormatTok->setBlockKind(BK_Block);
829 if (Tok->is(TT: TT_NamespaceLBrace))
830 FormatTok->setFinalizedType(TT_NamespaceRBrace);
831 }
832
833 const bool IsFunctionRBrace =
834 FormatTok->is(Kind: tok::r_brace) && Tok->is(TT: TT_FunctionLBrace);
835
836 auto RemoveBraces = [=]() mutable {
837 if (!SimpleBlock)
838 return false;
839 assert(Tok->isOneOf(TT_ControlStatementLBrace, TT_ElseLBrace));
840 assert(FormatTok->is(tok::r_brace));
841 const bool WrappedOpeningBrace = !Tok->Previous;
842 if (WrappedOpeningBrace && FollowedByComment)
843 return false;
844 const bool HasRequiredIfBraces = IfLBrace && !IfLBrace->Optional;
845 if (KeepBraces && !HasRequiredIfBraces)
846 return false;
847 if (Tok->isNot(Kind: TT_ElseLBrace) || !HasRequiredIfBraces) {
848 const FormatToken *Previous = Tokens->getPreviousToken();
849 assert(Previous);
850 if (Previous->is(Kind: tok::r_brace) && !Previous->Optional)
851 return false;
852 }
853 assert(!CurrentLines->empty());
854 auto &LastLine = CurrentLines->back();
855 if (LastLine.Level == InitialLevel + 1 && !mightFitOnOneLine(ParsedLine&: LastLine))
856 return false;
857 if (Tok->is(TT: TT_ElseLBrace))
858 return true;
859 if (WrappedOpeningBrace) {
860 assert(Index > 0);
861 --Index; // The line above the wrapped l_brace.
862 Tok = nullptr;
863 }
864 return mightFitOnOneLine(ParsedLine&: (*CurrentLines)[Index], OpeningBrace: Tok);
865 };
866 if (RemoveBraces()) {
867 Tok->MatchingParen = FormatTok;
868 FormatTok->MatchingParen = Tok;
869 }
870
871 size_t PPEndHash = computePPHash();
872
873 // Munch the closing brace.
874 nextToken(/*LevelDifference=*/-AddLevels);
875
876 // When this is a function block and there is an unnecessary semicolon
877 // afterwards then mark it as optional (so the RemoveSemi pass can get rid of
878 // it later).
879 if (Style.RemoveSemicolon && IsFunctionRBrace) {
880 while (FormatTok->is(Kind: tok::semi)) {
881 FormatTok->Optional = true;
882 nextToken();
883 }
884 }
885
886 HandleVerilogBlockLabel();
887
888 if (MacroBlock && FormatTok->is(Kind: tok::l_paren))
889 parseParens();
890
891 Line->Level = InitialLevel;
892
893 if (FormatTok->is(Kind: tok::kw_noexcept)) {
894 // A noexcept in a requires expression.
895 nextToken();
896 }
897
898 if (FormatTok->is(Kind: tok::arrow)) {
899 // Following the } or noexcept we can find a trailing return type arrow
900 // as part of an implicit conversion constraint.
901 nextToken();
902 parseStructuralElement();
903 }
904
905 if (MunchSemi && FormatTok->is(Kind: tok::semi))
906 nextToken();
907
908 if (PPStartHash == PPEndHash) {
909 Line->MatchingOpeningBlockLineIndex = OpeningLineIndex;
910 if (OpeningLineIndex != UnwrappedLine::kInvalidIndex) {
911 // Update the opening line to add the forward reference as well
912 (*CurrentLines)[OpeningLineIndex].MatchingClosingBlockLineIndex =
913 CurrentLines->size() - 1;
914 }
915 }
916
917 return IfLBrace;
918}
919
920static bool isGoogScope(const UnwrappedLine &Line) {
921 // FIXME: Closure-library specific stuff should not be hard-coded but be
922 // configurable.
923 if (Line.Tokens.size() < 4)
924 return false;
925 auto I = Line.Tokens.begin();
926 if (I->Tok->TokenText != "goog")
927 return false;
928 ++I;
929 if (I->Tok->isNot(Kind: tok::period))
930 return false;
931 ++I;
932 if (I->Tok->TokenText != "scope")
933 return false;
934 ++I;
935 return I->Tok->is(Kind: tok::l_paren);
936}
937
938static bool isIIFE(const UnwrappedLine &Line,
939 const AdditionalKeywords &Keywords) {
940 // Look for the start of an immediately invoked anonymous function.
941 // https://en.wikipedia.org/wiki/Immediately-invoked_function_expression
942 // This is commonly done in JavaScript to create a new, anonymous scope.
943 // Example: (function() { ... })()
944 if (Line.Tokens.size() < 3)
945 return false;
946 auto I = Line.Tokens.begin();
947 if (I->Tok->isNot(Kind: tok::l_paren))
948 return false;
949 ++I;
950 if (I->Tok->isNot(Kind: Keywords.kw_function))
951 return false;
952 ++I;
953 return I->Tok->is(Kind: tok::l_paren);
954}
955
956static bool ShouldBreakBeforeBrace(const FormatStyle &Style,
957 const FormatToken &InitialToken,
958 bool IsEmptyBlock,
959 bool IsJavaRecord = false) {
960 if (IsJavaRecord)
961 return Style.BraceWrapping.AfterClass;
962
963 tok::TokenKind Kind = InitialToken.Tok.getKind();
964 if (InitialToken.is(TT: TT_NamespaceMacro))
965 Kind = tok::kw_namespace;
966
967 const bool WrapRecordAllowed =
968 !IsEmptyBlock ||
969 Style.AllowShortRecordOnASingleLine < FormatStyle::SRS_Empty ||
970 Style.BraceWrapping.SplitEmptyRecord;
971
972 switch (Kind) {
973 case tok::kw_namespace:
974 return Style.BraceWrapping.AfterNamespace;
975 case tok::kw_class:
976 return Style.BraceWrapping.AfterClass && WrapRecordAllowed;
977 case tok::kw_union:
978 return Style.BraceWrapping.AfterUnion && WrapRecordAllowed;
979 case tok::kw_struct:
980 return Style.BraceWrapping.AfterStruct && WrapRecordAllowed;
981 case tok::kw_enum:
982 return Style.BraceWrapping.AfterEnum;
983 default:
984 return false;
985 }
986}
987
988void UnwrappedLineParser::parseChildBlock() {
989 assert(FormatTok->is(tok::l_brace));
990 FormatTok->setBlockKind(BK_Block);
991 const FormatToken *OpeningBrace = FormatTok;
992 nextToken();
993 {
994 bool SkipIndent = (Style.isJavaScript() &&
995 (isGoogScope(Line: *Line) || isIIFE(Line: *Line, Keywords)));
996 ScopedLineState LineState(*this);
997 ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack,
998 /*MustBeDeclaration=*/false);
999 Line->Level += SkipIndent ? 0 : 1;
1000 parseLevel(OpeningBrace);
1001 flushComments(NewlineBeforeNext: isOnNewLine(FormatTok: *FormatTok));
1002 Line->Level -= SkipIndent ? 0 : 1;
1003 }
1004 nextToken();
1005}
1006
1007void UnwrappedLineParser::parsePPDirective() {
1008 assert(FormatTok->is(tok::hash) && "'#' expected");
1009 ScopedMacroState MacroState(*Line, Tokens, FormatTok);
1010
1011 nextToken();
1012
1013 if (!FormatTok->Tok.getIdentifierInfo()) {
1014 parsePPUnknown();
1015 return;
1016 }
1017
1018 switch (FormatTok->Tok.getIdentifierInfo()->getPPKeywordID()) {
1019 case tok::pp_define:
1020 parsePPDefine();
1021 return;
1022 case tok::pp_if:
1023 parsePPIf(/*IfDef=*/false);
1024 break;
1025 case tok::pp_ifdef:
1026 case tok::pp_ifndef:
1027 parsePPIf(/*IfDef=*/true);
1028 break;
1029 case tok::pp_else:
1030 case tok::pp_elifdef:
1031 case tok::pp_elifndef:
1032 case tok::pp_elif:
1033 parsePPElse();
1034 break;
1035 case tok::pp_endif:
1036 parsePPEndIf();
1037 break;
1038 case tok::pp_pragma:
1039 parsePPPragma();
1040 break;
1041 case tok::pp_error:
1042 case tok::pp_warning:
1043 nextToken();
1044 if (!eof() && Style.isCpp())
1045 FormatTok->setFinalizedType(TT_AfterPPDirective);
1046 [[fallthrough]];
1047 default:
1048 parsePPUnknown();
1049 break;
1050 }
1051}
1052
1053void UnwrappedLineParser::conditionalCompilationCondition(bool Unreachable) {
1054 size_t Line = CurrentLines->size();
1055 if (CurrentLines == &PreprocessorDirectives)
1056 Line += Lines.size();
1057
1058 if (Unreachable ||
1059 (!PPStack.empty() && PPStack.back().Kind == PP_Unreachable)) {
1060 PPStack.push_back(Elt: {PP_Unreachable, Line});
1061 } else {
1062 PPStack.push_back(Elt: {PP_Conditional, Line});
1063 }
1064}
1065
1066void UnwrappedLineParser::conditionalCompilationStart(bool Unreachable) {
1067 ++PPBranchLevel;
1068 assert(PPBranchLevel >= 0 && PPBranchLevel <= (int)PPLevelBranchIndex.size());
1069 if (PPBranchLevel == (int)PPLevelBranchIndex.size()) {
1070 PPLevelBranchIndex.push_back(Elt: 0);
1071 PPLevelBranchCount.push_back(Elt: 0);
1072 }
1073 PPChainBranchIndex.push(x: Unreachable ? -1 : 0);
1074 bool Skip = PPLevelBranchIndex[PPBranchLevel] > 0;
1075 conditionalCompilationCondition(Unreachable: Unreachable || Skip);
1076}
1077
1078void UnwrappedLineParser::conditionalCompilationAlternative() {
1079 if (!PPStack.empty())
1080 PPStack.pop_back();
1081 assert(PPBranchLevel < (int)PPLevelBranchIndex.size());
1082 if (!PPChainBranchIndex.empty())
1083 ++PPChainBranchIndex.top();
1084 conditionalCompilationCondition(
1085 Unreachable: PPBranchLevel >= 0 && !PPChainBranchIndex.empty() &&
1086 PPLevelBranchIndex[PPBranchLevel] != PPChainBranchIndex.top());
1087}
1088
1089void UnwrappedLineParser::conditionalCompilationEnd() {
1090 assert(PPBranchLevel < (int)PPLevelBranchIndex.size());
1091 if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty()) {
1092 if (PPChainBranchIndex.top() + 1 > PPLevelBranchCount[PPBranchLevel])
1093 PPLevelBranchCount[PPBranchLevel] = PPChainBranchIndex.top() + 1;
1094 }
1095 // Guard against #endif's without #if.
1096 if (PPBranchLevel > -1)
1097 --PPBranchLevel;
1098 if (!PPChainBranchIndex.empty())
1099 PPChainBranchIndex.pop();
1100 if (!PPStack.empty())
1101 PPStack.pop_back();
1102}
1103
1104void UnwrappedLineParser::parsePPIf(bool IfDef) {
1105 bool IfNDef = FormatTok->is(Kind: tok::pp_ifndef);
1106 nextToken();
1107 bool Unreachable = false;
1108 if (!IfDef && (FormatTok->is(Kind: tok::kw_false) || FormatTok->TokenText == "0"))
1109 Unreachable = true;
1110 if (IfDef && !IfNDef && FormatTok->TokenText == "SWIG")
1111 Unreachable = true;
1112 conditionalCompilationStart(Unreachable);
1113 FormatToken *IfCondition = FormatTok;
1114 // If there's a #ifndef on the first line, and the only lines before it are
1115 // comments, it could be an include guard.
1116 bool MaybeIncludeGuard = IfNDef;
1117 if (IncludeGuard == IG_Inited && MaybeIncludeGuard) {
1118 for (auto &Line : Lines) {
1119 if (Line.Tokens.front().Tok->isNot(Kind: tok::comment)) {
1120 MaybeIncludeGuard = false;
1121 IncludeGuard = IG_Rejected;
1122 break;
1123 }
1124 }
1125 }
1126 --PPBranchLevel;
1127 parsePPUnknown();
1128 ++PPBranchLevel;
1129 if (IncludeGuard == IG_Inited && MaybeIncludeGuard) {
1130 IncludeGuard = IG_IfNdefed;
1131 IncludeGuardToken = IfCondition;
1132 }
1133}
1134
1135void UnwrappedLineParser::parsePPElse() {
1136 // If a potential include guard has an #else, it's not an include guard.
1137 if (IncludeGuard == IG_Defined && PPBranchLevel == 0)
1138 IncludeGuard = IG_Rejected;
1139 // Don't crash when there is an #else without an #if.
1140 assert(PPBranchLevel >= -1);
1141 if (PPBranchLevel == -1)
1142 conditionalCompilationStart(/*Unreachable=*/true);
1143 conditionalCompilationAlternative();
1144 --PPBranchLevel;
1145 parsePPUnknown();
1146 ++PPBranchLevel;
1147}
1148
1149void UnwrappedLineParser::parsePPEndIf() {
1150 conditionalCompilationEnd();
1151 parsePPUnknown();
1152}
1153
1154void UnwrappedLineParser::parsePPDefine() {
1155 nextToken();
1156
1157 if (!FormatTok->Tok.getIdentifierInfo()) {
1158 IncludeGuard = IG_Rejected;
1159 IncludeGuardToken = nullptr;
1160 parsePPUnknown();
1161 return;
1162 }
1163
1164 bool MaybeIncludeGuard = false;
1165 if (IncludeGuard == IG_IfNdefed &&
1166 IncludeGuardToken->TokenText == FormatTok->TokenText) {
1167 IncludeGuard = IG_Defined;
1168 IncludeGuardToken = nullptr;
1169 for (auto &Line : Lines) {
1170 if (Line.Tokens.front().Tok->isNoneOf(Ks: tok::comment, Ks: tok::hash)) {
1171 IncludeGuard = IG_Rejected;
1172 break;
1173 }
1174 }
1175 MaybeIncludeGuard = IncludeGuard == IG_Defined;
1176 }
1177
1178 // In the context of a define, even keywords should be treated as normal
1179 // identifiers. Setting the kind to identifier is not enough, because we need
1180 // to treat additional keywords like __except as well, which are already
1181 // identifiers. Setting the identifier info to null interferes with include
1182 // guard processing above, and changes preprocessing nesting.
1183 FormatTok->Tok.setKind(tok::identifier);
1184 FormatTok->Tok.setIdentifierInfo(Keywords.kw_internal_ident_after_define);
1185 nextToken();
1186
1187 // IncludeGuard can't have a non-empty macro definition.
1188 if (MaybeIncludeGuard && !eof())
1189 IncludeGuard = IG_Rejected;
1190
1191 if (FormatTok->is(Kind: tok::l_paren) && !FormatTok->hasWhitespaceBefore())
1192 parseParens();
1193 if (Style.IndentPPDirectives != FormatStyle::PPDIS_None)
1194 Line->Level += PPBranchLevel + 1;
1195 addUnwrappedLine();
1196 ++Line->Level;
1197
1198 Line->PPLevel = PPBranchLevel + (IncludeGuard == IG_Defined ? 0 : 1);
1199 assert((int)Line->PPLevel >= 0);
1200
1201 if (eof())
1202 return;
1203
1204 Line->InMacroBody = true;
1205
1206 if (!Style.SkipMacroDefinitionBody) {
1207 // Errors during a preprocessor directive can only affect the layout of the
1208 // preprocessor directive, and thus we ignore them. An alternative approach
1209 // would be to use the same approach we use on the file level (no
1210 // re-indentation if there was a structural error) within the macro
1211 // definition.
1212 parseFile();
1213 return;
1214 }
1215
1216 for (auto *Comment : CommentsBeforeNextToken)
1217 Comment->Finalized = true;
1218
1219 do {
1220 FormatTok->Finalized = true;
1221 FormatTok = Tokens->getNextToken();
1222 } while (!eof());
1223
1224 addUnwrappedLine();
1225}
1226
1227void UnwrappedLineParser::parsePPPragma() {
1228 Line->InPragmaDirective = true;
1229 parsePPUnknown();
1230}
1231
1232void UnwrappedLineParser::parsePPUnknown() {
1233 while (!eof())
1234 nextToken();
1235 if (Style.IndentPPDirectives != FormatStyle::PPDIS_None)
1236 Line->Level += PPBranchLevel + 1;
1237 addUnwrappedLine();
1238}
1239
1240// Here we exclude certain tokens that are not usually the first token in an
1241// unwrapped line. This is used in attempt to distinguish macro calls without
1242// trailing semicolons from other constructs split to several lines.
1243static bool tokenCanStartNewLine(const FormatToken &Tok) {
1244 // Semicolon can be a null-statement, l_square can be a start of a macro or
1245 // a C++11 attribute, but this doesn't seem to be common.
1246 return Tok.isNoneOf(Ks: tok::semi, Ks: tok::l_brace,
1247 // Tokens that can only be used as binary operators and a
1248 // part of overloaded operator names.
1249 Ks: tok::period, Ks: tok::periodstar, Ks: tok::arrow, Ks: tok::arrowstar,
1250 Ks: tok::less, Ks: tok::greater, Ks: tok::slash, Ks: tok::percent,
1251 Ks: tok::lessless, Ks: tok::greatergreater, Ks: tok::equal,
1252 Ks: tok::plusequal, Ks: tok::minusequal, Ks: tok::starequal,
1253 Ks: tok::slashequal, Ks: tok::percentequal, Ks: tok::ampequal,
1254 Ks: tok::pipeequal, Ks: tok::caretequal, Ks: tok::greatergreaterequal,
1255 Ks: tok::lesslessequal,
1256 // Colon is used in labels, base class lists, initializer
1257 // lists, range-based for loops, ternary operator, but
1258 // should never be the first token in an unwrapped line.
1259 Ks: tok::colon,
1260 // 'noexcept' is a trailing annotation.
1261 Ks: tok::kw_noexcept);
1262}
1263
1264static bool mustBeJSIdent(const AdditionalKeywords &Keywords,
1265 const FormatToken *FormatTok) {
1266 // FIXME: This returns true for C/C++ keywords like 'struct'.
1267 return FormatTok->is(Kind: tok::identifier) &&
1268 (!FormatTok->Tok.getIdentifierInfo() ||
1269 FormatTok->isNoneOf(
1270 Ks: Keywords.kw_in, Ks: Keywords.kw_of, Ks: Keywords.kw_as, Ks: Keywords.kw_async,
1271 Ks: Keywords.kw_await, Ks: Keywords.kw_yield, Ks: Keywords.kw_finally,
1272 Ks: Keywords.kw_function, Ks: Keywords.kw_import, Ks: Keywords.kw_is,
1273 Ks: Keywords.kw_let, Ks: Keywords.kw_var, Ks: tok::kw_const,
1274 Ks: Keywords.kw_abstract, Ks: Keywords.kw_extends, Ks: Keywords.kw_implements,
1275 Ks: Keywords.kw_instanceof, Ks: Keywords.kw_interface,
1276 Ks: Keywords.kw_override, Ks: Keywords.kw_throws, Ks: Keywords.kw_from));
1277}
1278
1279static bool mustBeJSIdentOrValue(const AdditionalKeywords &Keywords,
1280 const FormatToken *FormatTok) {
1281 return FormatTok->Tok.isLiteral() ||
1282 FormatTok->isOneOf(K1: tok::kw_true, K2: tok::kw_false) ||
1283 mustBeJSIdent(Keywords, FormatTok);
1284}
1285
1286// isJSDeclOrStmt returns true if |FormatTok| starts a declaration or statement
1287// when encountered after a value (see mustBeJSIdentOrValue).
1288static bool isJSDeclOrStmt(const AdditionalKeywords &Keywords,
1289 const FormatToken *FormatTok) {
1290 return FormatTok->isOneOf(
1291 K1: tok::kw_return, K2: Keywords.kw_yield,
1292 // conditionals
1293 Ks: tok::kw_if, Ks: tok::kw_else,
1294 // loops
1295 Ks: tok::kw_for, Ks: tok::kw_while, Ks: tok::kw_do, Ks: tok::kw_continue, Ks: tok::kw_break,
1296 // switch/case
1297 Ks: tok::kw_switch, Ks: tok::kw_case,
1298 // exceptions
1299 Ks: tok::kw_throw, Ks: tok::kw_try, Ks: tok::kw_catch, Ks: Keywords.kw_finally,
1300 // declaration
1301 Ks: tok::kw_const, Ks: tok::kw_class, Ks: Keywords.kw_var, Ks: Keywords.kw_let,
1302 Ks: Keywords.kw_async, Ks: Keywords.kw_function,
1303 // import/export
1304 Ks: Keywords.kw_import, Ks: tok::kw_export);
1305}
1306
1307// Checks whether a token is a type in K&R C (aka C78).
1308static bool isC78Type(const FormatToken &Tok) {
1309 return Tok.isOneOf(K1: tok::kw_char, K2: tok::kw_short, Ks: tok::kw_int, Ks: tok::kw_long,
1310 Ks: tok::kw_unsigned, Ks: tok::kw_float, Ks: tok::kw_double,
1311 Ks: tok::identifier);
1312}
1313
1314// This function checks whether a token starts the first parameter declaration
1315// in a K&R C (aka C78) function definition, e.g.:
1316// int f(a, b)
1317// short a, b;
1318// {
1319// return a + b;
1320// }
1321static bool isC78ParameterDecl(const FormatToken *Tok, const FormatToken *Next,
1322 const FormatToken *FuncName) {
1323 assert(Tok);
1324 assert(Next);
1325 assert(FuncName);
1326
1327 if (FuncName->isNot(Kind: tok::identifier))
1328 return false;
1329
1330 const FormatToken *Prev = FuncName->Previous;
1331 if (!Prev || (Prev->isNot(Kind: tok::star) && !isC78Type(Tok: *Prev)))
1332 return false;
1333
1334 if (!isC78Type(Tok: *Tok) &&
1335 Tok->isNoneOf(Ks: tok::kw_register, Ks: tok::kw_struct, Ks: tok::kw_union)) {
1336 return false;
1337 }
1338
1339 if (Next->isNot(Kind: tok::star) && !Next->Tok.getIdentifierInfo())
1340 return false;
1341
1342 Tok = Tok->Previous;
1343 if (!Tok || Tok->isNot(Kind: tok::r_paren))
1344 return false;
1345
1346 Tok = Tok->Previous;
1347 if (!Tok || Tok->isNot(Kind: tok::identifier))
1348 return false;
1349
1350 return Tok->Previous && Tok->Previous->isOneOf(K1: tok::l_paren, K2: tok::comma);
1351}
1352
1353bool UnwrappedLineParser::parseModuleDecl() {
1354 assert(IsCpp);
1355 assert(FormatTok->is(Keywords.kw_module));
1356
1357 if (Style.Language == FormatStyle::LK_C ||
1358 Style.Standard < FormatStyle::LS_Cpp20) {
1359 return false;
1360 }
1361
1362 nextToken();
1363 if (FormatTok->isNot(Kind: tok::identifier))
1364 return false;
1365
1366 for (nextToken(); FormatTok->isNoneOf(Ks: tok::semi, Ks: tok::eof); nextToken())
1367 if (FormatTok->is(Kind: tok::colon))
1368 FormatTok->setFinalizedType(TT_ModulePartitionColon);
1369
1370 nextToken();
1371 Line->IsModuleOrImportDecl = true;
1372 addUnwrappedLine();
1373 return true;
1374}
1375
1376bool UnwrappedLineParser::parseImportDecl() {
1377 assert(IsCpp);
1378 assert(FormatTok->is(Keywords.kw_import) && "'import' expected");
1379
1380 if (Style.Language == FormatStyle::LK_C ||
1381 Style.Standard < FormatStyle::LS_Cpp20) {
1382 return false;
1383 }
1384
1385 nextToken();
1386 if (FormatTok->is(Kind: tok::colon)) {
1387 FormatTok->setFinalizedType(TT_ModulePartitionColon);
1388 nextToken();
1389 }
1390 if (FormatTok->isNoneOf(Ks: tok::identifier, Ks: tok::less, Ks: tok::string_literal))
1391 return false;
1392
1393 for (; FormatTok->isNoneOf(Ks: tok::semi, Ks: tok::eof); nextToken()) {
1394 // Handle import <foo/bar.h> as we would an include statement.
1395 if (FormatTok->is(Kind: tok::less)) {
1396 for (nextToken(); FormatTok->isNoneOf(Ks: tok::greater, Ks: tok::semi, Ks: tok::eof);
1397 nextToken()) {
1398 // Mark tokens as implicit string literals, so that import <A/Foo> will
1399 // neither be broken nor have a space added.
1400 FormatTok->setFinalizedType(TT_ImplicitStringLiteral);
1401 }
1402 }
1403 }
1404
1405 nextToken();
1406 Line->IsModuleOrImportDecl = true;
1407 addUnwrappedLine();
1408 return true;
1409}
1410
1411// readTokenWithJavaScriptASI reads the next token and terminates the current
1412// line if JavaScript Automatic Semicolon Insertion must
1413// happen between the current token and the next token.
1414//
1415// This method is conservative - it cannot cover all edge cases of JavaScript,
1416// but only aims to correctly handle certain well known cases. It *must not*
1417// return true in speculative cases.
1418void UnwrappedLineParser::readTokenWithJavaScriptASI() {
1419 FormatToken *Previous = FormatTok;
1420 readToken();
1421 FormatToken *Next = FormatTok;
1422
1423 bool IsOnSameLine =
1424 CommentsBeforeNextToken.empty()
1425 ? Next->NewlinesBefore == 0
1426 : CommentsBeforeNextToken.front()->NewlinesBefore == 0;
1427 if (IsOnSameLine)
1428 return;
1429
1430 bool PreviousMustBeValue = mustBeJSIdentOrValue(Keywords, FormatTok: Previous);
1431 bool PreviousStartsTemplateExpr =
1432 Previous->is(TT: TT_TemplateString) && Previous->TokenText.ends_with(Suffix: "${");
1433 if (PreviousMustBeValue || Previous->is(Kind: tok::r_paren)) {
1434 // If the line contains an '@' sign, the previous token might be an
1435 // annotation, which can precede another identifier/value.
1436 bool HasAt = llvm::any_of(Range&: Line->Tokens, P: [](UnwrappedLineNode &LineNode) {
1437 return LineNode.Tok->is(Kind: tok::at);
1438 });
1439 if (HasAt)
1440 return;
1441 }
1442 if (Next->is(Kind: tok::exclaim) && PreviousMustBeValue)
1443 return addUnwrappedLine();
1444 bool NextMustBeValue = mustBeJSIdentOrValue(Keywords, FormatTok: Next);
1445 bool NextEndsTemplateExpr =
1446 Next->is(TT: TT_TemplateString) && Next->TokenText.starts_with(Prefix: "}");
1447 if (NextMustBeValue && !NextEndsTemplateExpr && !PreviousStartsTemplateExpr &&
1448 (PreviousMustBeValue ||
1449 Previous->isOneOf(K1: tok::r_square, K2: tok::r_paren, Ks: tok::plusplus,
1450 Ks: tok::minusminus))) {
1451 return addUnwrappedLine();
1452 }
1453 if ((PreviousMustBeValue || Previous->is(Kind: tok::r_paren)) &&
1454 isJSDeclOrStmt(Keywords, FormatTok: Next)) {
1455 return addUnwrappedLine();
1456 }
1457}
1458
1459void UnwrappedLineParser::parseStructuralElement(
1460 const FormatToken *OpeningBrace, IfStmtKind *IfKind,
1461 FormatToken **IfLeftBrace, bool *HasDoWhile, bool *HasLabel) {
1462 if (Style.isTableGen() && FormatTok->is(Kind: tok::pp_include)) {
1463 nextToken();
1464 if (FormatTok->is(Kind: tok::string_literal))
1465 nextToken();
1466 addUnwrappedLine();
1467 return;
1468 }
1469
1470 if (IsCpp) {
1471 while (FormatTok->is(Kind: tok::l_square) && handleCppAttributes()) {
1472 }
1473 } else if (Style.isVerilog()) {
1474 // Skip attributes.
1475 while (FormatTok->is(Kind: tok::l_paren) &&
1476 Tokens->peekNextToken()->is(Kind: tok::star)) {
1477 parseParens();
1478 }
1479 skipVerilogQualifiers();
1480 // Skip things that can exist before keywords like 'if' and 'case'.
1481 if (FormatTok->isOneOf(K1: Keywords.kw_priority, K2: Keywords.kw_unique,
1482 Ks: Keywords.kw_unique0)) {
1483 nextToken();
1484 }
1485
1486 if (Keywords.isVerilogStructuredProcedure(Tok: *FormatTok)) {
1487 parseForOrWhileLoop(/*HasParens=*/false);
1488 return;
1489 }
1490 if (FormatTok->isOneOf(K1: Keywords.kw_foreach, K2: Keywords.kw_repeat)) {
1491 parseForOrWhileLoop();
1492 return;
1493 }
1494 if (FormatTok->isOneOf(K1: tok::kw_restrict, K2: Keywords.kw_assert,
1495 Ks: Keywords.kw_assume, Ks: Keywords.kw_cover)) {
1496 parseIfThenElse(IfKind, /*KeepBraces=*/false, /*IsVerilogAssert=*/true);
1497 return;
1498 }
1499 }
1500
1501 // Tokens that only make sense at the beginning of a line.
1502 if (FormatTok->isAccessSpecifierKeyword()) {
1503 if (Style.isJava() || Style.isJavaScript() || Style.isCSharp())
1504 nextToken();
1505 else
1506 parseAccessSpecifier();
1507 return;
1508 }
1509 switch (FormatTok->Tok.getKind()) {
1510 case tok::kw_asm: {
1511 // Track whether to skip formatting inline asm by finalizing the tokens
1512 // in the block. Formatting is skipped inside of braces by default.
1513 // A style option could be added to also skip formatting inside parens.
1514 bool DoNotFormat = false;
1515 tok::TokenKind OpenType;
1516 tok::TokenKind CloseType;
1517 nextToken();
1518 while (FormatTok &&
1519 FormatTok->isOneOf(K1: tok::kw_volatile, K2: tok::kw_inline, Ks: tok::kw_goto)) {
1520 nextToken();
1521 }
1522 if (!FormatTok)
1523 break;
1524 if (FormatTok->is(Kind: tok::l_brace)) {
1525 FormatTok->setFinalizedType(TT_InlineASMBrace);
1526 OpenType = tok::l_brace;
1527 CloseType = tok::r_brace;
1528 DoNotFormat = true;
1529 } else if (FormatTok->is(Kind: tok::l_paren)) {
1530 OpenType = tok::l_paren;
1531 CloseType = tok::r_paren;
1532 FormatTok->setFinalizedType(TT_InlineASMParen);
1533 } else {
1534 break;
1535 }
1536 if (DoNotFormat) {
1537 FormatToken *OpenTok = FormatTok;
1538 int NestLevel = 0;
1539 nextToken();
1540 while (FormatTok && !eof()) {
1541 if (FormatTok->is(Kind: OpenType)) {
1542 ++NestLevel;
1543 } else if (FormatTok->is(Kind: CloseType)) {
1544 --NestLevel;
1545 if (NestLevel < 1) {
1546 FormatTok->setFinalizedType(OpenTok->getType());
1547 nextToken();
1548 addUnwrappedLine();
1549 break;
1550 }
1551 }
1552 FormatTok->Finalized = true;
1553 nextToken();
1554 }
1555 }
1556 break;
1557 }
1558 case tok::kw_namespace:
1559 parseNamespace();
1560 return;
1561 case tok::kw_if: {
1562 if (Style.isJavaScript() && Line->MustBeDeclaration) {
1563 // field/method declaration.
1564 break;
1565 }
1566 FormatToken *Tok = parseIfThenElse(IfKind);
1567 if (IfLeftBrace)
1568 *IfLeftBrace = Tok;
1569 return;
1570 }
1571 case tok::kw_for:
1572 case tok::kw_while:
1573 if (Style.isJavaScript() && Line->MustBeDeclaration) {
1574 // field/method declaration.
1575 break;
1576 }
1577 parseForOrWhileLoop();
1578 return;
1579 case tok::kw_do:
1580 if (Style.isJavaScript() && Line->MustBeDeclaration) {
1581 // field/method declaration.
1582 break;
1583 }
1584 parseDoWhile();
1585 if (HasDoWhile)
1586 *HasDoWhile = true;
1587 return;
1588 case tok::kw_switch:
1589 if (Style.isJavaScript() && Line->MustBeDeclaration) {
1590 // 'switch: string' field declaration.
1591 break;
1592 }
1593 parseSwitch(/*IsExpr=*/false);
1594 return;
1595 case tok::kw_default: {
1596 // In Verilog default along with other labels are handled in the next loop.
1597 if (Style.isVerilog())
1598 break;
1599 if (Style.isJavaScript() && Line->MustBeDeclaration) {
1600 // 'default: string' field declaration.
1601 break;
1602 }
1603 auto *Default = FormatTok;
1604 nextToken();
1605 if (FormatTok->is(Kind: tok::colon)) {
1606 FormatTok->setFinalizedType(TT_CaseLabelColon);
1607 parseLabel();
1608 return;
1609 }
1610 if (FormatTok->is(Kind: tok::arrow)) {
1611 FormatTok->setFinalizedType(TT_CaseLabelArrow);
1612 Default->setFinalizedType(TT_SwitchExpressionLabel);
1613 parseLabel();
1614 return;
1615 }
1616 // e.g. "default void f() {}" in a Java interface.
1617 break;
1618 }
1619 case tok::kw_case:
1620 // Proto: there are no switch/case statements.
1621 if (Style.Language == FormatStyle::LK_Proto) {
1622 nextToken();
1623 return;
1624 }
1625 if (Style.isVerilog()) {
1626 parseBlock();
1627 addUnwrappedLine();
1628 return;
1629 }
1630 if (Style.isJavaScript() && Line->MustBeDeclaration) {
1631 // 'case: string' field declaration.
1632 nextToken();
1633 break;
1634 }
1635 parseCaseLabel();
1636 return;
1637 case tok::kw_goto:
1638 nextToken();
1639 if (FormatTok->is(Kind: tok::kw_case))
1640 nextToken();
1641 break;
1642 case tok::kw_try:
1643 case tok::kw___try:
1644 if (Style.isJavaScript() && Line->MustBeDeclaration) {
1645 // field/method declaration.
1646 break;
1647 }
1648 parseTryCatch();
1649 return;
1650 case tok::kw_extern:
1651 if (Style.isVerilog()) {
1652 // In Verilog an extern module declaration looks like a start of module.
1653 // But there is no body and endmodule. So we handle it separately.
1654 parseVerilogExtern();
1655 return;
1656 }
1657 nextToken();
1658 if (FormatTok->is(Kind: tok::string_literal)) {
1659 nextToken();
1660 if (FormatTok->is(Kind: tok::l_brace)) {
1661 if (Style.BraceWrapping.AfterExternBlock)
1662 addUnwrappedLine();
1663 // Either we indent or for backwards compatibility we follow the
1664 // AfterExternBlock style.
1665 unsigned AddLevels =
1666 (Style.IndentExternBlock == FormatStyle::IEBS_Indent) ||
1667 (Style.BraceWrapping.AfterExternBlock &&
1668 Style.IndentExternBlock ==
1669 FormatStyle::IEBS_AfterExternBlock)
1670 ? 1u
1671 : 0u;
1672 parseBlock(/*MustBeDeclaration=*/true, AddLevels);
1673 addUnwrappedLine();
1674 return;
1675 }
1676 }
1677 break;
1678 case tok::kw_export:
1679 if (IsCpp) {
1680 nextToken();
1681 if (FormatTok->is(Kind: tok::kw_namespace)) {
1682 parseNamespace();
1683 return;
1684 }
1685 if (FormatTok->is(Kind: tok::l_brace)) {
1686 parseCppExportBlock();
1687 return;
1688 }
1689 if (FormatTok->is(II: Keywords.kw_module) && parseModuleDecl())
1690 return;
1691 if (FormatTok->is(II: Keywords.kw_import) && parseImportDecl())
1692 return;
1693 break;
1694 }
1695 if (Style.isJavaScript()) {
1696 parseJavaScriptEs6ImportExport();
1697 return;
1698 }
1699 if (Style.isVerilog()) {
1700 parseVerilogExtern();
1701 return;
1702 }
1703 break;
1704 case tok::kw_inline:
1705 nextToken();
1706 if (FormatTok->is(Kind: tok::kw_namespace)) {
1707 parseNamespace();
1708 return;
1709 }
1710 break;
1711 case tok::identifier:
1712 if (FormatTok->is(TT: TT_ForEachMacro)) {
1713 parseForOrWhileLoop();
1714 return;
1715 }
1716 if (FormatTok->is(TT: TT_MacroBlockBegin)) {
1717 parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u,
1718 /*MunchSemi=*/false);
1719 return;
1720 }
1721 if (FormatTok->is(II: Keywords.kw_import)) {
1722 if (IsCpp && parseImportDecl())
1723 return;
1724 if (Style.isJavaScript()) {
1725 parseJavaScriptEs6ImportExport();
1726 return;
1727 }
1728 if (Style.Language == FormatStyle::LK_Proto) {
1729 nextToken();
1730 if (FormatTok->is(Kind: tok::kw_public))
1731 nextToken();
1732 if (FormatTok->isNot(Kind: tok::string_literal))
1733 return;
1734 nextToken();
1735 if (FormatTok->is(Kind: tok::semi))
1736 nextToken();
1737 addUnwrappedLine();
1738 return;
1739 }
1740 if (Style.isVerilog()) {
1741 parseVerilogExtern();
1742 return;
1743 }
1744 }
1745 if (IsCpp) {
1746 if (FormatTok->is(II: Keywords.kw_module) && parseModuleDecl())
1747 return;
1748 if (FormatTok->isOneOf(K1: Keywords.kw_signals, K2: Keywords.kw_qsignals,
1749 Ks: Keywords.kw_slots, Ks: Keywords.kw_qslots)) {
1750 nextToken();
1751 if (FormatTok->is(Kind: tok::colon)) {
1752 nextToken();
1753 addUnwrappedLine();
1754 return;
1755 }
1756 }
1757 if (FormatTok->is(TT: TT_StatementMacro)) {
1758 parseStatementMacro();
1759 return;
1760 }
1761 if (FormatTok->is(TT: TT_NamespaceMacro)) {
1762 parseNamespace();
1763 return;
1764 }
1765 }
1766 // In Verilog labels can be any expression, so we don't do them here.
1767 // JS doesn't have macros, and within classes colons indicate fields, not
1768 // labels.
1769 // TableGen doesn't have labels.
1770 if (!Style.isJavaScript() && !Style.isVerilog() && !Style.isTableGen() &&
1771 Tokens->peekNextToken()->is(Kind: tok::colon) && !Line->MustBeDeclaration) {
1772 nextToken();
1773 if (!Line->InMacroBody || CurrentLines->size() > 1)
1774 Line->Tokens.begin()->Tok->MustBreakBefore = true;
1775 FormatTok->setFinalizedType(TT_GotoLabelColon);
1776 parseLabel(/*IsGotoLabel=*/true);
1777 if (HasLabel)
1778 *HasLabel = true;
1779 return;
1780 }
1781 if (Style.isJava() && FormatTok->is(II: Keywords.kw_record)) {
1782 parseRecord(/*ParseAsExpr=*/false, /*IsJavaRecord=*/true);
1783 addUnwrappedLine();
1784 return;
1785 }
1786 // In all other cases, parse the declaration.
1787 break;
1788 default:
1789 break;
1790 }
1791
1792 bool SeenEqual = false;
1793 for (const bool InRequiresExpression =
1794 OpeningBrace && OpeningBrace->isOneOf(K1: TT_RequiresExpressionLBrace,
1795 K2: TT_CompoundRequirementLBrace);
1796 !eof();) {
1797 const FormatToken *Previous = FormatTok->Previous;
1798 switch (FormatTok->Tok.getKind()) {
1799 case tok::at:
1800 nextToken();
1801 if (FormatTok->is(Kind: tok::l_brace)) {
1802 nextToken();
1803 parseBracedList();
1804 break;
1805 }
1806 if (Style.isJava() && FormatTok->is(II: Keywords.kw_interface)) {
1807 nextToken();
1808 break;
1809 }
1810 switch (bool IsAutoRelease = false; FormatTok->Tok.getObjCKeywordID()) {
1811 case tok::objc_public:
1812 case tok::objc_protected:
1813 case tok::objc_package:
1814 case tok::objc_private:
1815 return parseAccessSpecifier();
1816 case tok::objc_interface:
1817 case tok::objc_implementation:
1818 return parseObjCInterfaceOrImplementation();
1819 case tok::objc_protocol:
1820 if (parseObjCProtocol())
1821 return;
1822 break;
1823 case tok::objc_end:
1824 return; // Handled by the caller.
1825 case tok::objc_optional:
1826 case tok::objc_required:
1827 nextToken();
1828 addUnwrappedLine();
1829 return;
1830 case tok::objc_autoreleasepool:
1831 IsAutoRelease = true;
1832 [[fallthrough]];
1833 case tok::objc_synchronized:
1834 nextToken();
1835 if (!IsAutoRelease && FormatTok->is(Kind: tok::l_paren)) {
1836 // Skip synchronization object
1837 parseParens();
1838 }
1839 if (FormatTok->is(Kind: tok::l_brace)) {
1840 if (Style.BraceWrapping.AfterControlStatement ==
1841 FormatStyle::BWACS_Always) {
1842 addUnwrappedLine();
1843 }
1844 parseBlock();
1845 }
1846 addUnwrappedLine();
1847 return;
1848 case tok::objc_try:
1849 // This branch isn't strictly necessary (the kw_try case below would
1850 // do this too after the tok::at is parsed above). But be explicit.
1851 parseTryCatch();
1852 return;
1853 default:
1854 break;
1855 }
1856 break;
1857 case tok::kw_requires: {
1858 if (IsCpp) {
1859 bool ParsedClause = parseRequires(SeenEqual);
1860 if (ParsedClause)
1861 return;
1862 } else {
1863 nextToken();
1864 }
1865 break;
1866 }
1867 case tok::kw_enum:
1868 // Ignore if this is part of "template <enum ..." or "... -> enum" or
1869 // "template <..., enum ...>".
1870 if (Previous && Previous->isOneOf(K1: tok::less, K2: tok::arrow, Ks: tok::comma)) {
1871 nextToken();
1872 break;
1873 }
1874
1875 // parseEnum falls through and does not yet add an unwrapped line as an
1876 // enum definition can start a structural element.
1877 if (!parseEnum())
1878 break;
1879 // This only applies to C++ and Verilog.
1880 if (!IsCpp && !Style.isVerilog()) {
1881 addUnwrappedLine();
1882 return;
1883 }
1884 break;
1885 case tok::kw_typedef:
1886 nextToken();
1887 if (FormatTok->isOneOf(K1: Keywords.kw_NS_ENUM, K2: Keywords.kw_NS_OPTIONS,
1888 Ks: Keywords.kw_CF_ENUM, Ks: Keywords.kw_CF_OPTIONS,
1889 Ks: Keywords.kw_CF_CLOSED_ENUM,
1890 Ks: Keywords.kw_NS_CLOSED_ENUM)) {
1891 parseEnum();
1892 }
1893 break;
1894 case tok::kw_class:
1895 if (Style.isVerilog()) {
1896 parseBlock();
1897 addUnwrappedLine();
1898 return;
1899 }
1900 if (Style.isTableGen()) {
1901 // Do nothing special. In this case the l_brace becomes FunctionLBrace.
1902 // This is same as def and so on.
1903 nextToken();
1904 break;
1905 }
1906 [[fallthrough]];
1907 case tok::kw_struct:
1908 case tok::kw_union:
1909 if (parseStructLike())
1910 return;
1911 break;
1912 case tok::kw_decltype:
1913 nextToken();
1914 if (FormatTok->is(Kind: tok::l_paren)) {
1915 parseParens();
1916 if (FormatTok->Previous &&
1917 FormatTok->Previous->endsSequence(K1: tok::r_paren, Tokens: tok::kw_auto,
1918 Tokens: tok::l_paren)) {
1919 Line->SeenDecltypeAuto = true;
1920 }
1921 }
1922 break;
1923 case tok::period:
1924 nextToken();
1925 // In Java, classes have an implicit static member "class".
1926 if (Style.isJava() && FormatTok && FormatTok->is(Kind: tok::kw_class))
1927 nextToken();
1928 if (Style.isJavaScript() && FormatTok &&
1929 FormatTok->Tok.getIdentifierInfo()) {
1930 // JavaScript only has pseudo keywords, all keywords are allowed to
1931 // appear in "IdentifierName" positions. See http://es5.github.io/#x7.6
1932 nextToken();
1933 }
1934 break;
1935 case tok::semi:
1936 nextToken();
1937 addUnwrappedLine();
1938 return;
1939 case tok::r_brace:
1940 addUnwrappedLine();
1941 return;
1942 case tok::string_literal:
1943 if (Style.isVerilog() && FormatTok->is(TT: TT_VerilogProtected)) {
1944 FormatTok->Finalized = true;
1945 nextToken();
1946 addUnwrappedLine();
1947 return;
1948 }
1949 nextToken();
1950 break;
1951 case tok::l_paren: {
1952 parseParens();
1953 // Break the unwrapped line if a K&R C function definition has a parameter
1954 // declaration.
1955 if (OpeningBrace || !IsCpp || !Previous || eof())
1956 break;
1957 if (isC78ParameterDecl(Tok: FormatTok,
1958 Next: Tokens->peekNextToken(/*SkipComment=*/true),
1959 FuncName: Previous)) {
1960 addUnwrappedLine();
1961 return;
1962 }
1963 break;
1964 }
1965 case tok::kw_operator:
1966 nextToken();
1967 if (FormatTok->isBinaryOperator())
1968 nextToken();
1969 break;
1970 case tok::caret: {
1971 const auto *Prev = FormatTok->getPreviousNonComment();
1972 nextToken();
1973 if (Prev && Prev->is(Kind: tok::identifier))
1974 break;
1975 // Block return type.
1976 if (FormatTok->Tok.isAnyIdentifier() || FormatTok->isTypeName(LangOpts)) {
1977 nextToken();
1978 // Return types: pointers are ok too.
1979 while (FormatTok->is(Kind: tok::star))
1980 nextToken();
1981 }
1982 // Block argument list.
1983 if (FormatTok->is(Kind: tok::l_paren))
1984 parseParens();
1985 // Block body.
1986 if (FormatTok->is(Kind: tok::l_brace))
1987 parseChildBlock();
1988 break;
1989 }
1990 case tok::l_brace:
1991 if (InRequiresExpression)
1992 FormatTok->setFinalizedType(TT_BracedListLBrace);
1993 if (!tryToParsePropertyAccessor() && !tryToParseBracedList()) {
1994 IsDecltypeAutoFunction = Line->SeenDecltypeAuto;
1995 // A block outside of parentheses must be the last part of a
1996 // structural element.
1997 // FIXME: Figure out cases where this is not true, and add projections
1998 // for them (the one we know is missing are lambdas).
1999 if (Style.isJava() &&
2000 Line->Tokens.front().Tok->is(II: Keywords.kw_synchronized)) {
2001 // If necessary, we could set the type to something different than
2002 // TT_FunctionLBrace.
2003 if (Style.BraceWrapping.AfterControlStatement ==
2004 FormatStyle::BWACS_Always) {
2005 addUnwrappedLine();
2006 }
2007 } else if (Style.BraceWrapping.AfterFunction) {
2008 addUnwrappedLine();
2009 }
2010 if (!Previous || Previous->isNot(Kind: TT_TypeDeclarationParen))
2011 FormatTok->setFinalizedType(TT_FunctionLBrace);
2012 parseBlock();
2013 IsDecltypeAutoFunction = false;
2014 addUnwrappedLine();
2015 return;
2016 }
2017 // Otherwise this was a braced init list, and the structural
2018 // element continues.
2019 break;
2020 case tok::kw_try:
2021 if (Style.isJavaScript() && Line->MustBeDeclaration) {
2022 // field/method declaration.
2023 nextToken();
2024 break;
2025 }
2026 // We arrive here when parsing function-try blocks.
2027 if (Style.BraceWrapping.AfterFunction)
2028 addUnwrappedLine();
2029 parseTryCatch();
2030 return;
2031 case tok::identifier: {
2032 if (Style.isCSharp() && FormatTok->is(II: Keywords.kw_where) &&
2033 Line->MustBeDeclaration) {
2034 addUnwrappedLine();
2035 parseCSharpGenericTypeConstraint();
2036 break;
2037 }
2038 if (FormatTok->is(TT: TT_MacroBlockEnd)) {
2039 addUnwrappedLine();
2040 return;
2041 }
2042
2043 // Function declarations (as opposed to function expressions) are parsed
2044 // on their own unwrapped line by continuing this loop. Function
2045 // expressions (functions that are not on their own line) must not create
2046 // a new unwrapped line, so they are special cased below.
2047 size_t TokenCount = Line->Tokens.size();
2048 if (Style.isJavaScript() && FormatTok->is(II: Keywords.kw_function) &&
2049 (TokenCount > 1 ||
2050 (TokenCount == 1 &&
2051 Line->Tokens.front().Tok->isNot(Kind: Keywords.kw_async)))) {
2052 tryToParseJSFunction();
2053 break;
2054 }
2055 if ((Style.isJavaScript() || Style.isJava()) &&
2056 FormatTok->is(II: Keywords.kw_interface)) {
2057 if (Style.isJavaScript()) {
2058 // In JavaScript/TypeScript, "interface" can be used as a standalone
2059 // identifier, e.g. in `var interface = 1;`. If "interface" is
2060 // followed by another identifier, it is very like to be an actual
2061 // interface declaration.
2062 unsigned StoredPosition = Tokens->getPosition();
2063 FormatToken *Next = Tokens->getNextToken();
2064 FormatTok = Tokens->setPosition(StoredPosition);
2065 if (!mustBeJSIdent(Keywords, FormatTok: Next)) {
2066 nextToken();
2067 break;
2068 }
2069 }
2070 parseRecord();
2071 addUnwrappedLine();
2072 return;
2073 }
2074
2075 if (Style.isVerilog()) {
2076 if (FormatTok->is(II: Keywords.kw_table)) {
2077 parseVerilogTable();
2078 return;
2079 }
2080 if (Keywords.isVerilogBegin(Tok: *FormatTok) ||
2081 Keywords.isVerilogHierarchy(Tok: *FormatTok)) {
2082 parseBlock();
2083 addUnwrappedLine();
2084 return;
2085 }
2086 }
2087
2088 if (!IsCpp && FormatTok->is(II: Keywords.kw_interface)) {
2089 if (parseStructLike())
2090 return;
2091 break;
2092 }
2093
2094 if (IsCpp && FormatTok->is(TT: TT_StatementMacro)) {
2095 parseStatementMacro();
2096 return;
2097 }
2098
2099 // See if the following token should start a new unwrapped line.
2100 StringRef Text = FormatTok->TokenText;
2101
2102 FormatToken *PreviousToken = FormatTok;
2103 nextToken();
2104
2105 // JS doesn't have macros, and within classes colons indicate fields, not
2106 // labels.
2107 if (Style.isJavaScript())
2108 break;
2109
2110 auto OneTokenSoFar = [&]() {
2111 auto I = Line->Tokens.begin(), E = Line->Tokens.end();
2112 while (I != E && I->Tok->is(Kind: tok::comment))
2113 ++I;
2114 if (Style.isVerilog())
2115 while (I != E && I->Tok->is(Kind: tok::hash))
2116 ++I;
2117 return I != E && (++I == E);
2118 };
2119 if (OneTokenSoFar()) {
2120 // Recognize function-like macro usages without trailing semicolon as
2121 // well as free-standing macros like Q_OBJECT.
2122 bool FunctionLike = FormatTok->is(Kind: tok::l_paren);
2123 if (FunctionLike)
2124 parseParens();
2125
2126 bool FollowedByNewline =
2127 CommentsBeforeNextToken.empty()
2128 ? FormatTok->NewlinesBefore > 0
2129 : CommentsBeforeNextToken.front()->NewlinesBefore > 0;
2130
2131 if (FollowedByNewline &&
2132 (Text.size() >= 5 ||
2133 (FunctionLike && FormatTok->isNot(Kind: tok::l_paren))) &&
2134 tokenCanStartNewLine(Tok: *FormatTok) && Text == Text.upper()) {
2135 if (PreviousToken->isNot(Kind: TT_UntouchableMacroFunc))
2136 PreviousToken->setFinalizedType(TT_FunctionLikeOrFreestandingMacro);
2137 addUnwrappedLine();
2138 return;
2139 }
2140 }
2141 break;
2142 }
2143 case tok::equal:
2144 if ((Style.isJavaScript() || Style.isCSharp()) &&
2145 FormatTok->is(TT: TT_FatArrow)) {
2146 tryToParseChildBlock();
2147 break;
2148 }
2149
2150 SeenEqual = true;
2151 nextToken();
2152 if (FormatTok->is(Kind: tok::l_brace)) {
2153 // C# needs this change to ensure that array initialisers and object
2154 // initialisers are indented the same way. In TypeScript, the brace
2155 // can also be an object type definition.
2156 if (!Style.isJavaScript())
2157 FormatTok->setBlockKind(BK_BracedInit);
2158 // TableGen's defset statement has syntax of the form,
2159 // `defset <type> <name> = { <statement>... }`
2160 if (Style.isTableGen() &&
2161 Line->Tokens.begin()->Tok->is(II: Keywords.kw_defset)) {
2162 FormatTok->setFinalizedType(TT_FunctionLBrace);
2163 parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u,
2164 /*MunchSemi=*/false);
2165 addUnwrappedLine();
2166 break;
2167 }
2168 nextToken();
2169 parseBracedList();
2170 } else if (Style.Language == FormatStyle::LK_Proto &&
2171 FormatTok->is(Kind: tok::less)) {
2172 nextToken();
2173 parseBracedList(/*IsAngleBracket=*/true);
2174 }
2175 break;
2176 case tok::l_square:
2177 parseSquare();
2178 break;
2179 case tok::kw_new:
2180 if (Style.isCSharp() &&
2181 (Tokens->peekNextToken()->isAccessSpecifierKeyword() ||
2182 (Previous && Previous->isAccessSpecifierKeyword()))) {
2183 nextToken();
2184 } else {
2185 parseNew();
2186 }
2187 break;
2188 case tok::kw_switch:
2189 if (Style.isJava())
2190 parseSwitch(/*IsExpr=*/true);
2191 else
2192 nextToken();
2193 break;
2194 case tok::kw_case:
2195 // Proto: there are no switch/case statements.
2196 if (Style.Language == FormatStyle::LK_Proto) {
2197 nextToken();
2198 return;
2199 }
2200 // In Verilog switch is called case.
2201 if (Style.isVerilog()) {
2202 parseBlock();
2203 addUnwrappedLine();
2204 return;
2205 }
2206 if (Style.isJavaScript() && Line->MustBeDeclaration) {
2207 // 'case: string' field declaration.
2208 nextToken();
2209 break;
2210 }
2211 parseCaseLabel();
2212 break;
2213 case tok::kw_default:
2214 nextToken();
2215 if (Style.isVerilog()) {
2216 if (FormatTok->is(Kind: tok::colon)) {
2217 // The label will be handled in the next iteration.
2218 break;
2219 }
2220 if (FormatTok->is(II: Keywords.kw_clocking)) {
2221 // A default clocking block.
2222 parseBlock();
2223 addUnwrappedLine();
2224 return;
2225 }
2226 parseVerilogCaseLabel();
2227 return;
2228 }
2229 break;
2230 case tok::colon:
2231 nextToken();
2232 if (Style.isVerilog()) {
2233 parseVerilogCaseLabel();
2234 return;
2235 }
2236 break;
2237 case tok::greater:
2238 nextToken();
2239 if (FormatTok->is(Kind: tok::l_brace))
2240 FormatTok->Previous->setFinalizedType(TT_TemplateCloser);
2241 break;
2242 default:
2243 nextToken();
2244 break;
2245 }
2246 }
2247}
2248
2249bool UnwrappedLineParser::tryToParsePropertyAccessor() {
2250 assert(FormatTok->is(tok::l_brace));
2251 if (!Style.isCSharp())
2252 return false;
2253 // See if it's a property accessor.
2254 if (!FormatTok->Previous || FormatTok->Previous->isNot(Kind: tok::identifier))
2255 return false;
2256
2257 // See if we are inside a property accessor.
2258 //
2259 // Record the current tokenPosition so that we can advance and
2260 // reset the current token. `Next` is not set yet so we need
2261 // another way to advance along the token stream.
2262 unsigned int StoredPosition = Tokens->getPosition();
2263 FormatToken *Tok = Tokens->getNextToken();
2264
2265 // A trivial property accessor is of the form:
2266 // { [ACCESS_SPECIFIER] [get]; [ACCESS_SPECIFIER] [set|init] }
2267 // Track these as they do not require line breaks to be introduced.
2268 bool HasSpecialAccessor = false;
2269 bool IsTrivialPropertyAccessor = true;
2270 bool HasAttribute = false;
2271 while (!eof()) {
2272 if (const bool IsAccessorKeyword =
2273 Tok->isOneOf(K1: Keywords.kw_get, K2: Keywords.kw_init, Ks: Keywords.kw_set);
2274 IsAccessorKeyword || Tok->isAccessSpecifierKeyword() ||
2275 Tok->isOneOf(K1: tok::l_square, K2: tok::semi, Ks: Keywords.kw_internal)) {
2276 if (IsAccessorKeyword)
2277 HasSpecialAccessor = true;
2278 else if (Tok->is(Kind: tok::l_square))
2279 HasAttribute = true;
2280 Tok = Tokens->getNextToken();
2281 continue;
2282 }
2283 if (Tok->isNot(Kind: tok::r_brace))
2284 IsTrivialPropertyAccessor = false;
2285 break;
2286 }
2287
2288 if (!HasSpecialAccessor || HasAttribute) {
2289 Tokens->setPosition(StoredPosition);
2290 return false;
2291 }
2292
2293 // Try to parse the property accessor:
2294 // https://docs.microsoft.com/en-us/dotnet/csharp/programming-guide/classes-and-structs/properties
2295 Tokens->setPosition(StoredPosition);
2296 if (!IsTrivialPropertyAccessor && Style.BraceWrapping.AfterFunction)
2297 addUnwrappedLine();
2298 nextToken();
2299 do {
2300 switch (FormatTok->Tok.getKind()) {
2301 case tok::r_brace:
2302 nextToken();
2303 if (FormatTok->is(Kind: tok::equal)) {
2304 while (!eof() && FormatTok->isNot(Kind: tok::semi))
2305 nextToken();
2306 nextToken();
2307 }
2308 addUnwrappedLine();
2309 return true;
2310 case tok::l_brace:
2311 ++Line->Level;
2312 parseBlock(/*MustBeDeclaration=*/true);
2313 addUnwrappedLine();
2314 --Line->Level;
2315 break;
2316 case tok::equal:
2317 if (FormatTok->is(TT: TT_FatArrow)) {
2318 ++Line->Level;
2319 do {
2320 nextToken();
2321 } while (!eof() && FormatTok->isNot(Kind: tok::semi));
2322 nextToken();
2323 addUnwrappedLine();
2324 --Line->Level;
2325 break;
2326 }
2327 nextToken();
2328 break;
2329 default:
2330 if (FormatTok->isOneOf(K1: Keywords.kw_get, K2: Keywords.kw_init,
2331 Ks: Keywords.kw_set) &&
2332 !IsTrivialPropertyAccessor) {
2333 // Non-trivial get/set needs to be on its own line.
2334 addUnwrappedLine();
2335 }
2336 nextToken();
2337 }
2338 } while (!eof());
2339
2340 // Unreachable for well-formed code (paired '{' and '}').
2341 return true;
2342}
2343
2344bool UnwrappedLineParser::tryToParseLambda() {
2345 assert(FormatTok->is(tok::l_square));
2346 if (!IsCpp) {
2347 nextToken();
2348 return false;
2349 }
2350 FormatToken &LSquare = *FormatTok;
2351 if (!tryToParseLambdaIntroducer())
2352 return false;
2353
2354 FormatToken *Arrow = nullptr;
2355 bool InTemplateParameterList = false;
2356
2357 while (FormatTok->isNot(Kind: tok::l_brace)) {
2358 if (FormatTok->isTypeName(LangOpts) || FormatTok->isAttribute()) {
2359 nextToken();
2360 continue;
2361 }
2362 switch (FormatTok->Tok.getKind()) {
2363 case tok::l_brace:
2364 break;
2365 case tok::l_paren:
2366 parseParens(/*AmpAmpTokenType=*/StarAndAmpTokenType: TT_PointerOrReference);
2367 break;
2368 case tok::l_square:
2369 parseSquare();
2370 break;
2371 case tok::less:
2372 assert(FormatTok->Previous);
2373 if (FormatTok->Previous->is(Kind: tok::r_square))
2374 InTemplateParameterList = true;
2375 nextToken();
2376 break;
2377 case tok::kw_auto:
2378 case tok::kw_class:
2379 case tok::kw_struct:
2380 case tok::kw_union:
2381 case tok::kw_template:
2382 case tok::kw_typename:
2383 case tok::amp:
2384 case tok::star:
2385 case tok::kw_const:
2386 case tok::kw_constexpr:
2387 case tok::kw_consteval:
2388 case tok::comma:
2389 case tok::greater:
2390 case tok::identifier:
2391 case tok::numeric_constant:
2392 case tok::coloncolon:
2393 case tok::kw_mutable:
2394 case tok::kw_noexcept:
2395 case tok::kw_static:
2396 nextToken();
2397 break;
2398 // Specialization of a template with an integer parameter can contain
2399 // arithmetic, logical, comparison and ternary operators.
2400 //
2401 // FIXME: This also accepts sequences of operators that are not in the scope
2402 // of a template argument list.
2403 //
2404 // In a C++ lambda a template type can only occur after an arrow. We use
2405 // this as an heuristic to distinguish between Objective-C expressions
2406 // followed by an `a->b` expression, such as:
2407 // ([obj func:arg] + a->b)
2408 // Otherwise the code below would parse as a lambda.
2409 case tok::plus:
2410 case tok::minus:
2411 case tok::exclaim:
2412 case tok::tilde:
2413 case tok::slash:
2414 case tok::percent:
2415 case tok::lessless:
2416 case tok::pipe:
2417 case tok::pipepipe:
2418 case tok::ampamp:
2419 case tok::caret:
2420 case tok::equalequal:
2421 case tok::exclaimequal:
2422 case tok::greaterequal:
2423 case tok::lessequal:
2424 case tok::question:
2425 case tok::colon:
2426 case tok::ellipsis:
2427 case tok::kw_true:
2428 case tok::kw_false:
2429 if (Arrow || InTemplateParameterList) {
2430 nextToken();
2431 break;
2432 }
2433 return true;
2434 case tok::arrow:
2435 Arrow = FormatTok;
2436 nextToken();
2437 break;
2438 case tok::kw_requires:
2439 parseRequiresClause();
2440 break;
2441 case tok::equal:
2442 if (!InTemplateParameterList)
2443 return true;
2444 nextToken();
2445 break;
2446 default:
2447 return true;
2448 }
2449 }
2450
2451 FormatTok->setFinalizedType(TT_LambdaLBrace);
2452 LSquare.setFinalizedType(TT_LambdaLSquare);
2453
2454 if (Arrow)
2455 Arrow->setFinalizedType(TT_LambdaArrow);
2456
2457 NestedLambdas.push_back(Elt: Line->SeenDecltypeAuto);
2458 parseChildBlock();
2459 assert(!NestedLambdas.empty());
2460 NestedLambdas.pop_back();
2461
2462 return true;
2463}
2464
2465bool UnwrappedLineParser::tryToParseLambdaIntroducer() {
2466 const FormatToken *Previous = FormatTok->Previous;
2467 const FormatToken *LeftSquare = FormatTok;
2468 nextToken();
2469 if (Previous) {
2470 const auto *PrevPrev = Previous->getPreviousNonComment();
2471 if (Previous->is(Kind: tok::star) && PrevPrev && PrevPrev->isTypeName(LangOpts))
2472 return false;
2473 if (Previous->closesScope()) {
2474 // Not a potential C-style cast.
2475 if (Previous->isNot(Kind: tok::r_paren))
2476 return false;
2477 // Lambdas can be cast to function types only, e.g. `std::function<int()>`
2478 // and `int (*)()`.
2479 if (!PrevPrev || PrevPrev->isNoneOf(Ks: tok::greater, Ks: tok::r_paren))
2480 return false;
2481 }
2482 if (Previous && Previous->Tok.getIdentifierInfo() &&
2483 Previous->isNoneOf(Ks: tok::kw_return, Ks: tok::kw_co_await, Ks: tok::kw_co_yield,
2484 Ks: tok::kw_co_return)) {
2485 return false;
2486 }
2487 }
2488 if (LeftSquare->isCppStructuredBinding(IsCpp))
2489 return false;
2490 if (FormatTok->is(Kind: tok::l_square) || tok::isLiteral(K: FormatTok->Tok.getKind()))
2491 return false;
2492 if (FormatTok->is(Kind: tok::r_square)) {
2493 const FormatToken *Next = Tokens->peekNextToken(/*SkipComment=*/true);
2494 if (Next->is(Kind: tok::greater))
2495 return false;
2496 }
2497 parseSquare(/*LambdaIntroducer=*/true);
2498 return true;
2499}
2500
2501void UnwrappedLineParser::tryToParseJSFunction() {
2502 assert(FormatTok->is(Keywords.kw_function));
2503 if (FormatTok->is(II: Keywords.kw_async))
2504 nextToken();
2505 // Consume "function".
2506 nextToken();
2507
2508 // Consume * (generator function). Treat it like C++'s overloaded operators.
2509 if (FormatTok->is(Kind: tok::star)) {
2510 FormatTok->setFinalizedType(TT_OverloadedOperator);
2511 nextToken();
2512 }
2513
2514 // Consume function name.
2515 if (FormatTok->is(Kind: tok::identifier))
2516 nextToken();
2517
2518 if (FormatTok->isNot(Kind: tok::l_paren))
2519 return;
2520
2521 // Parse formal parameter list.
2522 parseParens();
2523
2524 if (FormatTok->is(Kind: tok::colon)) {
2525 // Parse a type definition.
2526 nextToken();
2527
2528 // Eat the type declaration. For braced inline object types, balance braces,
2529 // otherwise just parse until finding an l_brace for the function body.
2530 if (FormatTok->is(Kind: tok::l_brace))
2531 tryToParseBracedList();
2532 else
2533 while (FormatTok->isNoneOf(Ks: tok::l_brace, Ks: tok::semi) && !eof())
2534 nextToken();
2535 }
2536
2537 if (FormatTok->is(Kind: tok::semi))
2538 return;
2539
2540 parseChildBlock();
2541}
2542
2543bool UnwrappedLineParser::tryToParseBracedList() {
2544 if (FormatTok->is(BBK: BK_Unknown))
2545 calculateBraceTypes();
2546 assert(FormatTok->isNot(BK_Unknown));
2547 if (FormatTok->is(BBK: BK_Block))
2548 return false;
2549 nextToken();
2550 parseBracedList();
2551 return true;
2552}
2553
2554bool UnwrappedLineParser::tryToParseChildBlock() {
2555 assert(Style.isJavaScript() || Style.isCSharp());
2556 assert(FormatTok->is(TT_FatArrow));
2557 // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType TT_FatArrow.
2558 // They always start an expression or a child block if followed by a curly
2559 // brace.
2560 nextToken();
2561 if (FormatTok->isNot(Kind: tok::l_brace))
2562 return false;
2563 parseChildBlock();
2564 return true;
2565}
2566
2567bool UnwrappedLineParser::parseBracedList(bool IsAngleBracket, bool IsEnum) {
2568 assert(!IsAngleBracket || !IsEnum);
2569 bool HasError = false;
2570
2571 // FIXME: Once we have an expression parser in the UnwrappedLineParser,
2572 // replace this by using parseAssignmentExpression() inside.
2573 do {
2574 if (Style.isCSharp() && FormatTok->is(TT: TT_FatArrow) &&
2575 tryToParseChildBlock()) {
2576 continue;
2577 }
2578 if (Style.isJavaScript()) {
2579 if (FormatTok->is(II: Keywords.kw_function)) {
2580 tryToParseJSFunction();
2581 continue;
2582 }
2583 if (FormatTok->is(Kind: tok::l_brace)) {
2584 // Could be a method inside of a braced list `{a() { return 1; }}`.
2585 if (tryToParseBracedList())
2586 continue;
2587 parseChildBlock();
2588 }
2589 }
2590 if (FormatTok->is(Kind: IsAngleBracket ? tok::greater : tok::r_brace)) {
2591 if (IsEnum) {
2592 FormatTok->setBlockKind(BK_Block);
2593 if (!Style.AllowShortEnumsOnASingleLine)
2594 addUnwrappedLine();
2595 }
2596 nextToken();
2597 return !HasError;
2598 }
2599 switch (FormatTok->Tok.getKind()) {
2600 case tok::l_square:
2601 if (Style.isCSharp())
2602 parseSquare();
2603 else
2604 tryToParseLambda();
2605 break;
2606 case tok::l_paren:
2607 parseParens();
2608 // JavaScript can just have free standing methods and getters/setters in
2609 // object literals. Detect them by a "{" following ")".
2610 if (Style.isJavaScript()) {
2611 if (FormatTok->is(Kind: tok::l_brace))
2612 parseChildBlock();
2613 break;
2614 }
2615 break;
2616 case tok::l_brace:
2617 // Assume there are no blocks inside a braced init list apart
2618 // from the ones we explicitly parse out (like lambdas).
2619 FormatTok->setBlockKind(BK_BracedInit);
2620 if (!IsAngleBracket) {
2621 auto *Prev = FormatTok->Previous;
2622 if (Prev && Prev->is(Kind: tok::greater))
2623 Prev->setFinalizedType(TT_TemplateCloser);
2624 }
2625 nextToken();
2626 parseBracedList();
2627 break;
2628 case tok::less:
2629 nextToken();
2630 if (IsAngleBracket)
2631 parseBracedList(/*IsAngleBracket=*/true);
2632 break;
2633 case tok::semi:
2634 // JavaScript (or more precisely TypeScript) can have semicolons in braced
2635 // lists (in so-called TypeMemberLists). Thus, the semicolon cannot be
2636 // used for error recovery if we have otherwise determined that this is
2637 // a braced list.
2638 if (Style.isJavaScript()) {
2639 nextToken();
2640 break;
2641 }
2642 HasError = true;
2643 if (!IsEnum)
2644 return false;
2645 nextToken();
2646 break;
2647 case tok::comma:
2648 nextToken();
2649 if (IsEnum && !Style.AllowShortEnumsOnASingleLine)
2650 addUnwrappedLine();
2651 break;
2652 case tok::kw_requires:
2653 parseRequiresExpression();
2654 break;
2655 default:
2656 nextToken();
2657 break;
2658 }
2659 } while (!eof());
2660 return false;
2661}
2662
2663/// Parses a pair of parentheses (and everything between them).
2664/// \param StarAndAmpTokenType If different than TT_Unknown sets this type for
2665/// all (double) ampersands and stars. This applies for all nested scopes as
2666/// well, this is disabled within a (potential) template argument <>, and thus
2667/// also if we find only a <.
2668///
2669/// Returns whether there is a `=` token between the parentheses.
2670bool UnwrappedLineParser::parseParens(TokenType StarAndAmpTokenType,
2671 bool InMacroCall) {
2672 assert(FormatTok->is(tok::l_paren) && "'(' expected.");
2673 auto *LParen = FormatTok;
2674 auto *Prev = FormatTok->Previous;
2675 bool SeenComma = false;
2676 bool SeenEqual = false;
2677 bool MightBeFoldExpr = false;
2678 unsigned ExcessLess = 0;
2679 nextToken();
2680 const bool MightBeStmtExpr = FormatTok->is(Kind: tok::l_brace);
2681 if (!InMacroCall && Prev && Prev->is(TT: TT_FunctionLikeMacro))
2682 InMacroCall = true;
2683 do {
2684 switch (FormatTok->Tok.getKind()) {
2685 case tok::l_paren:
2686 if (parseParens(StarAndAmpTokenType: ExcessLess == 0 ? StarAndAmpTokenType : TT_Unknown,
2687 InMacroCall)) {
2688 SeenEqual = true;
2689 }
2690 if (Style.isJava() && FormatTok->is(Kind: tok::l_brace))
2691 parseChildBlock();
2692 break;
2693 case tok::r_paren: {
2694 auto *RParen = FormatTok;
2695 nextToken();
2696 if (Prev) {
2697 auto OptionalParens = [&] {
2698 if (Style.RemoveParentheses == FormatStyle::RPS_Leave ||
2699 MightBeStmtExpr || MightBeFoldExpr || SeenComma || InMacroCall ||
2700 Line->InMacroBody || RParen->getPreviousNonComment() == LParen) {
2701 return false;
2702 }
2703 const bool DoubleParens =
2704 Prev->is(Kind: tok::l_paren) && FormatTok->is(Kind: tok::r_paren);
2705 if (DoubleParens) {
2706 const auto *PrevPrev = Prev->getPreviousNonComment();
2707 const bool Excluded =
2708 PrevPrev &&
2709 (PrevPrev->isOneOf(K1: tok::kw___attribute, K2: tok::kw_decltype) ||
2710 (SeenEqual &&
2711 (PrevPrev->isOneOf(K1: tok::kw_if, K2: tok::kw_while) ||
2712 PrevPrev->endsSequence(K1: tok::kw_constexpr, Tokens: tok::kw_if))));
2713 if (!Excluded)
2714 return true;
2715 } else {
2716 const bool CommaSeparated =
2717 Prev->isOneOf(K1: tok::l_paren, K2: tok::comma) &&
2718 FormatTok->isOneOf(K1: tok::comma, K2: tok::r_paren);
2719 if (CommaSeparated &&
2720 // LParen is not preceded by ellipsis, comma.
2721 !Prev->endsSequence(K1: tok::comma, Tokens: tok::ellipsis) &&
2722 // RParen is not followed by comma, ellipsis.
2723 !(FormatTok->is(Kind: tok::comma) &&
2724 Tokens->peekNextToken()->is(Kind: tok::ellipsis))) {
2725 return true;
2726 }
2727 const bool ReturnParens =
2728 Style.RemoveParentheses == FormatStyle::RPS_ReturnStatement &&
2729 ((NestedLambdas.empty() && !IsDecltypeAutoFunction) ||
2730 (!NestedLambdas.empty() && !NestedLambdas.back())) &&
2731 Prev->isOneOf(K1: tok::kw_return, K2: tok::kw_co_return) &&
2732 FormatTok->is(Kind: tok::semi);
2733 if (ReturnParens)
2734 return true;
2735 }
2736 return false;
2737 };
2738 if (OptionalParens()) {
2739 LParen->Optional = true;
2740 RParen->Optional = true;
2741 } else if (Prev->is(TT: TT_TypenameMacro)) {
2742 LParen->setFinalizedType(TT_TypeDeclarationParen);
2743 RParen->setFinalizedType(TT_TypeDeclarationParen);
2744 } else if (Prev->is(Kind: tok::greater) && RParen->Previous == LParen) {
2745 Prev->setFinalizedType(TT_TemplateCloser);
2746 } else if (FormatTok->is(Kind: tok::l_brace) && Prev->is(Kind: tok::amp) &&
2747 !Prev->Previous) {
2748 FormatTok->setBlockKind(BK_BracedInit);
2749 }
2750 }
2751 return SeenEqual;
2752 }
2753 case tok::r_brace:
2754 // A "}" inside parenthesis is an error if there wasn't a matching "{".
2755 return SeenEqual;
2756 case tok::l_square:
2757 tryToParseLambda();
2758 break;
2759 case tok::l_brace:
2760 if (!tryToParseBracedList())
2761 parseChildBlock();
2762 break;
2763 case tok::at:
2764 nextToken();
2765 if (FormatTok->is(Kind: tok::l_brace)) {
2766 nextToken();
2767 parseBracedList();
2768 }
2769 break;
2770 case tok::comma:
2771 SeenComma = true;
2772 nextToken();
2773 break;
2774 case tok::ellipsis:
2775 MightBeFoldExpr = true;
2776 nextToken();
2777 break;
2778 case tok::equal:
2779 SeenEqual = true;
2780 if (Style.isCSharp() && FormatTok->is(TT: TT_FatArrow))
2781 tryToParseChildBlock();
2782 else
2783 nextToken();
2784 break;
2785 case tok::kw_class:
2786 if (Style.isJavaScript())
2787 parseRecord(/*ParseAsExpr=*/true);
2788 else
2789 nextToken();
2790 break;
2791 case tok::identifier:
2792 if (Style.isJavaScript() && (FormatTok->is(II: Keywords.kw_function)))
2793 tryToParseJSFunction();
2794 else
2795 nextToken();
2796 break;
2797 case tok::kw_switch:
2798 if (Style.isJava())
2799 parseSwitch(/*IsExpr=*/true);
2800 else
2801 nextToken();
2802 break;
2803 case tok::kw_requires:
2804 parseRequiresExpression();
2805 break;
2806 case tok::less:
2807 // We have here no clue whether this is a less, or a template opener, opt
2808 // out of the predefined StarAndAmpTokenType.
2809 ++ExcessLess;
2810 nextToken();
2811 break;
2812 case tok::greater:
2813 if (ExcessLess > 0)
2814 --ExcessLess;
2815 nextToken();
2816 break;
2817 case tok::star:
2818 case tok::amp:
2819 case tok::ampamp:
2820 if (StarAndAmpTokenType != TT_Unknown && ExcessLess == 0)
2821 FormatTok->setFinalizedType(StarAndAmpTokenType);
2822 [[fallthrough]];
2823 default:
2824 nextToken();
2825 break;
2826 }
2827 } while (!eof());
2828 return SeenEqual;
2829}
2830
2831void UnwrappedLineParser::parseSquare(bool LambdaIntroducer) {
2832 if (!LambdaIntroducer) {
2833 assert(FormatTok->is(tok::l_square) && "'[' expected.");
2834 if (tryToParseLambda())
2835 return;
2836 }
2837 do {
2838 switch (FormatTok->Tok.getKind()) {
2839 case tok::l_paren:
2840 parseParens();
2841 break;
2842 case tok::r_square:
2843 nextToken();
2844 return;
2845 case tok::r_brace:
2846 // A "}" inside parenthesis is an error if there wasn't a matching "{".
2847 return;
2848 case tok::l_square:
2849 parseSquare();
2850 break;
2851 case tok::l_brace: {
2852 if (!tryToParseBracedList())
2853 parseChildBlock();
2854 break;
2855 }
2856 case tok::at:
2857 case tok::colon:
2858 nextToken();
2859 if (FormatTok->is(Kind: tok::l_brace)) {
2860 nextToken();
2861 parseBracedList();
2862 }
2863 break;
2864 default:
2865 nextToken();
2866 break;
2867 }
2868 } while (!eof());
2869}
2870
2871void UnwrappedLineParser::keepAncestorBraces() {
2872 if (!Style.RemoveBracesLLVM)
2873 return;
2874
2875 const int MaxNestingLevels = 2;
2876 const int Size = NestedTooDeep.size();
2877 if (Size >= MaxNestingLevels)
2878 NestedTooDeep[Size - MaxNestingLevels] = true;
2879 NestedTooDeep.push_back(Elt: false);
2880}
2881
2882static FormatToken *getLastNonComment(const UnwrappedLine &Line) {
2883 for (const auto &Token : llvm::reverse(C: Line.Tokens))
2884 if (Token.Tok->isNot(Kind: tok::comment))
2885 return Token.Tok;
2886
2887 return nullptr;
2888}
2889
2890void UnwrappedLineParser::parseUnbracedBody(bool CheckEOF) {
2891 FormatToken *Tok = nullptr;
2892
2893 if (Style.InsertBraces && !Line->InPPDirective && !Line->Tokens.empty() &&
2894 PreprocessorDirectives.empty() && FormatTok->isNot(Kind: tok::semi)) {
2895 Tok = Style.BraceWrapping.AfterControlStatement == FormatStyle::BWACS_Never
2896 ? getLastNonComment(Line: *Line)
2897 : Line->Tokens.back().Tok;
2898 assert(Tok);
2899 if (Tok->BraceCount < 0) {
2900 assert(Tok->BraceCount == -1);
2901 Tok = nullptr;
2902 } else {
2903 Tok->BraceCount = -1;
2904 }
2905 }
2906
2907 addUnwrappedLine();
2908 ++Line->Level;
2909 ++Line->UnbracedBodyLevel;
2910 parseStructuralElement();
2911 --Line->UnbracedBodyLevel;
2912
2913 if (Tok) {
2914 assert(!Line->InPPDirective);
2915 Tok = nullptr;
2916 for (const auto &L : llvm::reverse(C&: *CurrentLines)) {
2917 if (!L.InPPDirective && getLastNonComment(Line: L)) {
2918 Tok = L.Tokens.back().Tok;
2919 break;
2920 }
2921 }
2922 assert(Tok);
2923 ++Tok->BraceCount;
2924 }
2925
2926 if (CheckEOF && eof())
2927 addUnwrappedLine();
2928
2929 --Line->Level;
2930}
2931
2932static void markOptionalBraces(FormatToken *LeftBrace) {
2933 if (!LeftBrace)
2934 return;
2935
2936 assert(LeftBrace->is(tok::l_brace));
2937
2938 FormatToken *RightBrace = LeftBrace->MatchingParen;
2939 if (!RightBrace) {
2940 assert(!LeftBrace->Optional);
2941 return;
2942 }
2943
2944 assert(RightBrace->is(tok::r_brace));
2945 assert(RightBrace->MatchingParen == LeftBrace);
2946 assert(LeftBrace->Optional == RightBrace->Optional);
2947
2948 LeftBrace->Optional = true;
2949 RightBrace->Optional = true;
2950}
2951
2952void UnwrappedLineParser::handleAttributes() {
2953 // Handle AttributeMacro, e.g. `if (x) UNLIKELY`.
2954 if (FormatTok->isAttribute())
2955 nextToken();
2956 else if (FormatTok->is(Kind: tok::l_square))
2957 handleCppAttributes();
2958}
2959
2960bool UnwrappedLineParser::handleCppAttributes() {
2961 // Handle [[likely]] / [[unlikely]] attributes.
2962 assert(FormatTok->is(tok::l_square));
2963 if (!tryToParseSimpleAttribute())
2964 return false;
2965 parseSquare();
2966 return true;
2967}
2968
2969/// Returns whether \c Tok begins a block.
2970bool UnwrappedLineParser::isBlockBegin(const FormatToken &Tok) const {
2971 // FIXME: rename the function or make
2972 // Tok.isOneOf(tok::l_brace, TT_MacroBlockBegin) work.
2973 return Style.isVerilog() ? Keywords.isVerilogBegin(Tok)
2974 : Tok.is(Kind: tok::l_brace);
2975}
2976
2977FormatToken *UnwrappedLineParser::parseIfThenElse(IfStmtKind *IfKind,
2978 bool KeepBraces,
2979 bool IsVerilogAssert) {
2980 assert((FormatTok->is(tok::kw_if) ||
2981 (Style.isVerilog() &&
2982 FormatTok->isOneOf(tok::kw_restrict, Keywords.kw_assert,
2983 Keywords.kw_assume, Keywords.kw_cover))) &&
2984 "'if' expected");
2985 nextToken();
2986
2987 if (IsVerilogAssert) {
2988 // Handle `assert #0` and `assert final`.
2989 if (FormatTok->is(II: Keywords.kw_verilogHash)) {
2990 nextToken();
2991 if (FormatTok->is(Kind: tok::numeric_constant))
2992 nextToken();
2993 } else if (FormatTok->isOneOf(K1: Keywords.kw_final, K2: Keywords.kw_property,
2994 Ks: Keywords.kw_sequence)) {
2995 nextToken();
2996 }
2997 }
2998
2999 // TableGen's if statement has the form of `if <cond> then { ... }`.
3000 if (Style.isTableGen()) {
3001 while (!eof() && FormatTok->isNot(Kind: Keywords.kw_then)) {
3002 // Simply skip until then. This range only contains a value.
3003 nextToken();
3004 }
3005 }
3006
3007 // Handle `if !consteval`.
3008 if (FormatTok->is(Kind: tok::exclaim))
3009 nextToken();
3010
3011 bool KeepIfBraces = true;
3012 if (FormatTok->is(Kind: tok::kw_consteval)) {
3013 nextToken();
3014 } else {
3015 KeepIfBraces = !Style.RemoveBracesLLVM || KeepBraces;
3016 if (FormatTok->isOneOf(K1: tok::kw_constexpr, K2: tok::identifier))
3017 nextToken();
3018 if (FormatTok->is(Kind: tok::l_paren)) {
3019 FormatTok->setFinalizedType(TT_ConditionLParen);
3020 parseParens();
3021 }
3022 }
3023 handleAttributes();
3024 // The then action is optional in Verilog assert statements.
3025 if (IsVerilogAssert && FormatTok->is(Kind: tok::semi)) {
3026 nextToken();
3027 addUnwrappedLine();
3028 return nullptr;
3029 }
3030
3031 bool NeedsUnwrappedLine = false;
3032 keepAncestorBraces();
3033
3034 FormatToken *IfLeftBrace = nullptr;
3035 IfStmtKind IfBlockKind = IfStmtKind::NotIf;
3036
3037 if (isBlockBegin(Tok: *FormatTok)) {
3038 FormatTok->setFinalizedType(TT_ControlStatementLBrace);
3039 IfLeftBrace = FormatTok;
3040 CompoundStatementIndenter Indenter(this, Style, Line->Level);
3041 parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u,
3042 /*MunchSemi=*/true, KeepBraces: KeepIfBraces, IfKind: &IfBlockKind);
3043 setPreviousRBraceType(TT_ControlStatementRBrace);
3044 if (Style.BraceWrapping.BeforeElse)
3045 addUnwrappedLine();
3046 else
3047 NeedsUnwrappedLine = true;
3048 } else if (IsVerilogAssert && FormatTok->is(Kind: tok::kw_else)) {
3049 addUnwrappedLine();
3050 } else {
3051 parseUnbracedBody();
3052 }
3053
3054 if (Style.RemoveBracesLLVM) {
3055 assert(!NestedTooDeep.empty());
3056 KeepIfBraces = KeepIfBraces ||
3057 (IfLeftBrace && !IfLeftBrace->MatchingParen) ||
3058 NestedTooDeep.back() || IfBlockKind == IfStmtKind::IfOnly ||
3059 IfBlockKind == IfStmtKind::IfElseIf;
3060 }
3061
3062 bool KeepElseBraces = KeepIfBraces;
3063 FormatToken *ElseLeftBrace = nullptr;
3064 IfStmtKind Kind = IfStmtKind::IfOnly;
3065
3066 if (FormatTok->is(Kind: tok::kw_else)) {
3067 if (Style.RemoveBracesLLVM) {
3068 NestedTooDeep.back() = false;
3069 Kind = IfStmtKind::IfElse;
3070 }
3071 nextToken();
3072 handleAttributes();
3073 if (isBlockBegin(Tok: *FormatTok)) {
3074 const bool FollowedByIf = Tokens->peekNextToken()->is(Kind: tok::kw_if);
3075 FormatTok->setFinalizedType(TT_ElseLBrace);
3076 ElseLeftBrace = FormatTok;
3077 CompoundStatementIndenter Indenter(this, Style, Line->Level);
3078 IfStmtKind ElseBlockKind = IfStmtKind::NotIf;
3079 FormatToken *IfLBrace =
3080 parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u,
3081 /*MunchSemi=*/true, KeepBraces: KeepElseBraces, IfKind: &ElseBlockKind);
3082 setPreviousRBraceType(TT_ElseRBrace);
3083 if (FormatTok->is(Kind: tok::kw_else)) {
3084 KeepElseBraces = KeepElseBraces ||
3085 ElseBlockKind == IfStmtKind::IfOnly ||
3086 ElseBlockKind == IfStmtKind::IfElseIf;
3087 } else if (FollowedByIf && IfLBrace && !IfLBrace->Optional) {
3088 KeepElseBraces = true;
3089 assert(ElseLeftBrace->MatchingParen);
3090 markOptionalBraces(LeftBrace: ElseLeftBrace);
3091 }
3092 addUnwrappedLine();
3093 } else if (!IsVerilogAssert && FormatTok->is(Kind: tok::kw_if)) {
3094 const FormatToken *Previous = Tokens->getPreviousToken();
3095 assert(Previous);
3096 const bool IsPrecededByComment = Previous->is(Kind: tok::comment);
3097 if (IsPrecededByComment) {
3098 addUnwrappedLine();
3099 ++Line->Level;
3100 }
3101 bool TooDeep = true;
3102 if (Style.RemoveBracesLLVM) {
3103 Kind = IfStmtKind::IfElseIf;
3104 TooDeep = NestedTooDeep.pop_back_val();
3105 }
3106 ElseLeftBrace = parseIfThenElse(/*IfKind=*/nullptr, KeepBraces: KeepIfBraces);
3107 if (Style.RemoveBracesLLVM)
3108 NestedTooDeep.push_back(Elt: TooDeep);
3109 if (IsPrecededByComment)
3110 --Line->Level;
3111 } else {
3112 parseUnbracedBody(/*CheckEOF=*/true);
3113 }
3114 } else {
3115 KeepIfBraces = KeepIfBraces || IfBlockKind == IfStmtKind::IfElse;
3116 if (NeedsUnwrappedLine)
3117 addUnwrappedLine();
3118 }
3119
3120 if (!Style.RemoveBracesLLVM)
3121 return nullptr;
3122
3123 assert(!NestedTooDeep.empty());
3124 KeepElseBraces = KeepElseBraces ||
3125 (ElseLeftBrace && !ElseLeftBrace->MatchingParen) ||
3126 NestedTooDeep.back();
3127
3128 NestedTooDeep.pop_back();
3129
3130 if (!KeepIfBraces && !KeepElseBraces) {
3131 markOptionalBraces(LeftBrace: IfLeftBrace);
3132 markOptionalBraces(LeftBrace: ElseLeftBrace);
3133 } else if (IfLeftBrace) {
3134 FormatToken *IfRightBrace = IfLeftBrace->MatchingParen;
3135 if (IfRightBrace) {
3136 assert(IfRightBrace->MatchingParen == IfLeftBrace);
3137 assert(!IfLeftBrace->Optional);
3138 assert(!IfRightBrace->Optional);
3139 IfLeftBrace->MatchingParen = nullptr;
3140 IfRightBrace->MatchingParen = nullptr;
3141 }
3142 }
3143
3144 if (IfKind)
3145 *IfKind = Kind;
3146
3147 return IfLeftBrace;
3148}
3149
3150void UnwrappedLineParser::parseTryCatch() {
3151 assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected");
3152 nextToken();
3153 bool NeedsUnwrappedLine = false;
3154 bool HasCtorInitializer = false;
3155 if (FormatTok->is(Kind: tok::colon)) {
3156 auto *Colon = FormatTok;
3157 // We are in a function try block, what comes is an initializer list.
3158 nextToken();
3159 if (FormatTok->is(Kind: tok::identifier)) {
3160 HasCtorInitializer = true;
3161 Colon->setFinalizedType(TT_CtorInitializerColon);
3162 }
3163
3164 // In case identifiers were removed by clang-tidy, what might follow is
3165 // multiple commas in sequence - before the first identifier.
3166 while (FormatTok->is(Kind: tok::comma))
3167 nextToken();
3168
3169 while (FormatTok->is(Kind: tok::identifier)) {
3170 nextToken();
3171 if (FormatTok->is(Kind: tok::l_paren)) {
3172 parseParens();
3173 } else if (FormatTok->is(Kind: tok::l_brace)) {
3174 nextToken();
3175 parseBracedList();
3176 }
3177
3178 // In case identifiers were removed by clang-tidy, what might follow is
3179 // multiple commas in sequence - after the first identifier.
3180 while (FormatTok->is(Kind: tok::comma))
3181 nextToken();
3182 }
3183 }
3184 // Parse try with resource.
3185 if (Style.isJava() && FormatTok->is(Kind: tok::l_paren))
3186 parseParens();
3187
3188 keepAncestorBraces();
3189
3190 if (FormatTok->is(Kind: tok::l_brace)) {
3191 if (HasCtorInitializer)
3192 FormatTok->setFinalizedType(TT_FunctionLBrace);
3193 CompoundStatementIndenter Indenter(this, Style, Line->Level);
3194 parseBlock();
3195 if (Style.BraceWrapping.BeforeCatch)
3196 addUnwrappedLine();
3197 else
3198 NeedsUnwrappedLine = true;
3199 } else if (FormatTok->isNot(Kind: tok::kw_catch)) {
3200 // The C++ standard requires a compound-statement after a try.
3201 // If there's none, we try to assume there's a structuralElement
3202 // and try to continue.
3203 addUnwrappedLine();
3204 ++Line->Level;
3205 parseStructuralElement();
3206 --Line->Level;
3207 }
3208 for (bool SeenCatch = false;;) {
3209 if (FormatTok->is(Kind: tok::at))
3210 nextToken();
3211 if (FormatTok->isNoneOf(Ks: tok::kw_catch, Ks: Keywords.kw___except,
3212 Ks: tok::kw___finally, Ks: tok::objc_catch,
3213 Ks: tok::objc_finally) &&
3214 !((Style.isJava() || Style.isJavaScript()) &&
3215 FormatTok->is(II: Keywords.kw_finally))) {
3216 break;
3217 }
3218 if (FormatTok->is(Kind: tok::kw_catch))
3219 SeenCatch = true;
3220 nextToken();
3221 while (FormatTok->isNot(Kind: tok::l_brace)) {
3222 if (FormatTok->is(Kind: tok::l_paren)) {
3223 parseParens();
3224 continue;
3225 }
3226 if (FormatTok->isOneOf(K1: tok::semi, K2: tok::r_brace) || eof()) {
3227 if (Style.RemoveBracesLLVM)
3228 NestedTooDeep.pop_back();
3229 return;
3230 }
3231 nextToken();
3232 }
3233 if (SeenCatch) {
3234 FormatTok->setFinalizedType(TT_ControlStatementLBrace);
3235 SeenCatch = false;
3236 }
3237 NeedsUnwrappedLine = false;
3238 Line->MustBeDeclaration = false;
3239 CompoundStatementIndenter Indenter(this, Style, Line->Level);
3240 parseBlock();
3241 if (Style.BraceWrapping.BeforeCatch)
3242 addUnwrappedLine();
3243 else
3244 NeedsUnwrappedLine = true;
3245 }
3246
3247 if (Style.RemoveBracesLLVM)
3248 NestedTooDeep.pop_back();
3249
3250 if (NeedsUnwrappedLine)
3251 addUnwrappedLine();
3252}
3253
3254void UnwrappedLineParser::parseNamespaceOrExportBlock(unsigned AddLevels) {
3255 bool ManageWhitesmithsBraces =
3256 AddLevels == 0u && Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths;
3257
3258 // If we're in Whitesmiths mode, indent the brace if we're not indenting
3259 // the whole block.
3260 if (ManageWhitesmithsBraces)
3261 ++Line->Level;
3262
3263 // Munch the semicolon after the block. This is more common than one would
3264 // think. Putting the semicolon into its own line is very ugly.
3265 parseBlock(/*MustBeDeclaration=*/true, AddLevels, /*MunchSemi=*/true,
3266 /*KeepBraces=*/true, /*IfKind=*/nullptr, UnindentWhitesmithsBraces: ManageWhitesmithsBraces);
3267
3268 addUnwrappedLine(AdjustLevel: AddLevels > 0 ? LineLevel::Remove : LineLevel::Keep);
3269
3270 if (ManageWhitesmithsBraces)
3271 --Line->Level;
3272}
3273
3274void UnwrappedLineParser::parseNamespace() {
3275 assert(FormatTok->isOneOf(tok::kw_namespace, TT_NamespaceMacro) &&
3276 "'namespace' expected");
3277
3278 const FormatToken &InitialToken = *FormatTok;
3279 nextToken();
3280 if (InitialToken.is(TT: TT_NamespaceMacro)) {
3281 parseParens();
3282 } else {
3283 while (FormatTok->isOneOf(K1: tok::identifier, K2: tok::coloncolon, Ks: tok::kw_inline,
3284 Ks: tok::l_square, Ks: tok::period, Ks: tok::l_paren) ||
3285 (Style.isCSharp() && FormatTok->is(Kind: tok::kw_union))) {
3286 if (FormatTok->is(Kind: tok::l_square))
3287 parseSquare();
3288 else if (FormatTok->is(Kind: tok::l_paren))
3289 parseParens();
3290 else
3291 nextToken();
3292 }
3293 }
3294 if (FormatTok->is(Kind: tok::l_brace)) {
3295 FormatTok->setFinalizedType(TT_NamespaceLBrace);
3296
3297 if (ShouldBreakBeforeBrace(Style, InitialToken,
3298 IsEmptyBlock: Tokens->peekNextToken()->is(Kind: tok::r_brace))) {
3299 addUnwrappedLine();
3300 }
3301
3302 unsigned AddLevels =
3303 Style.NamespaceIndentation == FormatStyle::NI_All ||
3304 (Style.NamespaceIndentation == FormatStyle::NI_Inner &&
3305 DeclarationScopeStack.size() > 1)
3306 ? 1u
3307 : 0u;
3308 parseNamespaceOrExportBlock(AddLevels);
3309 }
3310 // FIXME: Add error handling.
3311}
3312
3313void UnwrappedLineParser::parseCppExportBlock() {
3314 if (FormatTok->is(Kind: tok::l_brace)) {
3315 FormatTok->setFinalizedType(TT_ExportLBrace);
3316 if (Style.BraceWrapping.AfterExportBlock)
3317 addUnwrappedLine();
3318 }
3319 parseNamespaceOrExportBlock(/*AddLevels=*/Style.IndentExportBlock ? 1 : 0);
3320}
3321
3322void UnwrappedLineParser::parseNew() {
3323 assert(FormatTok->is(tok::kw_new) && "'new' expected");
3324 nextToken();
3325
3326 if (Style.isCSharp()) {
3327 do {
3328 // Handle constructor invocation, e.g. `new(field: value)`.
3329 if (FormatTok->is(Kind: tok::l_paren))
3330 parseParens();
3331
3332 // Handle array initialization syntax, e.g. `new[] {10, 20, 30}`.
3333 if (FormatTok->is(Kind: tok::l_brace))
3334 parseBracedList();
3335
3336 if (FormatTok->isOneOf(K1: tok::semi, K2: tok::comma))
3337 return;
3338
3339 nextToken();
3340 } while (!eof());
3341 }
3342
3343 if (!Style.isJava())
3344 return;
3345
3346 // In Java, we can parse everything up to the parens, which aren't optional.
3347 do {
3348 // There should not be a ;, { or } before the new's open paren.
3349 if (FormatTok->isOneOf(K1: tok::semi, K2: tok::l_brace, Ks: tok::r_brace))
3350 return;
3351
3352 // Consume the parens.
3353 if (FormatTok->is(Kind: tok::l_paren)) {
3354 parseParens();
3355
3356 // If there is a class body of an anonymous class, consume that as child.
3357 if (FormatTok->is(Kind: tok::l_brace))
3358 parseChildBlock();
3359 return;
3360 }
3361 nextToken();
3362 } while (!eof());
3363}
3364
3365void UnwrappedLineParser::parseLoopBody(bool KeepBraces, bool WrapRightBrace) {
3366 keepAncestorBraces();
3367
3368 if (isBlockBegin(Tok: *FormatTok)) {
3369 FormatTok->setFinalizedType(TT_ControlStatementLBrace);
3370 FormatToken *LeftBrace = FormatTok;
3371 CompoundStatementIndenter Indenter(this, Style, Line->Level);
3372 parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u,
3373 /*MunchSemi=*/true, KeepBraces);
3374 setPreviousRBraceType(TT_ControlStatementRBrace);
3375 if (!KeepBraces) {
3376 assert(!NestedTooDeep.empty());
3377 if (!NestedTooDeep.back())
3378 markOptionalBraces(LeftBrace);
3379 }
3380 if (WrapRightBrace)
3381 addUnwrappedLine();
3382 } else {
3383 parseUnbracedBody();
3384 }
3385
3386 if (!KeepBraces)
3387 NestedTooDeep.pop_back();
3388}
3389
3390void UnwrappedLineParser::parseForOrWhileLoop(bool HasParens) {
3391 assert((FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) ||
3392 (Style.isVerilog() &&
3393 FormatTok->isOneOf(Keywords.kw_always, Keywords.kw_always_comb,
3394 Keywords.kw_always_ff, Keywords.kw_always_latch,
3395 Keywords.kw_final, Keywords.kw_initial,
3396 Keywords.kw_foreach, Keywords.kw_forever,
3397 Keywords.kw_repeat))) &&
3398 "'for', 'while' or foreach macro expected");
3399 const bool KeepBraces = !Style.RemoveBracesLLVM ||
3400 FormatTok->isNoneOf(Ks: tok::kw_for, Ks: tok::kw_while);
3401
3402 nextToken();
3403 // JS' for await ( ...
3404 if (Style.isJavaScript() && FormatTok->is(II: Keywords.kw_await))
3405 nextToken();
3406 if (IsCpp && FormatTok->is(Kind: tok::kw_co_await))
3407 nextToken();
3408 if (HasParens && FormatTok->is(Kind: tok::l_paren)) {
3409 // The type is only set for Verilog basically because we were afraid to
3410 // change the existing behavior for loops. See the discussion on D121756 for
3411 // details.
3412 if (Style.isVerilog())
3413 FormatTok->setFinalizedType(TT_ConditionLParen);
3414 parseParens();
3415 }
3416
3417 if (Style.isVerilog()) {
3418 // Event control.
3419 parseVerilogSensitivityList();
3420 } else if (Style.AllowShortLoopsOnASingleLine && FormatTok->is(Kind: tok::semi) &&
3421 Tokens->getPreviousToken()->is(Kind: tok::r_paren)) {
3422 nextToken();
3423 addUnwrappedLine();
3424 return;
3425 }
3426
3427 handleAttributes();
3428 parseLoopBody(KeepBraces, /*WrapRightBrace=*/true);
3429}
3430
3431void UnwrappedLineParser::parseDoWhile() {
3432 assert(FormatTok->is(tok::kw_do) && "'do' expected");
3433 nextToken();
3434
3435 parseLoopBody(/*KeepBraces=*/true, WrapRightBrace: Style.BraceWrapping.BeforeWhile);
3436
3437 // FIXME: Add error handling.
3438 if (FormatTok->isNot(Kind: tok::kw_while)) {
3439 addUnwrappedLine();
3440 return;
3441 }
3442
3443 FormatTok->setFinalizedType(TT_DoWhile);
3444
3445 // If in Whitesmiths mode, the line with the while() needs to be indented
3446 // to the same level as the block.
3447 if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths)
3448 ++Line->Level;
3449
3450 nextToken();
3451 parseStructuralElement();
3452}
3453
3454void UnwrappedLineParser::parseLabel(bool IsGotoLabel) {
3455 nextToken();
3456
3457 const auto IndentGotoLabel = Style.IndentGotoLabels;
3458 const auto OldLineLevel = Line->Level;
3459 auto &Level = Line->Level;
3460
3461 if (IsGotoLabel && IndentGotoLabel == FormatStyle::IGLS_NoIndent)
3462 Level = 0;
3463
3464 if (!IsGotoLabel || IndentGotoLabel == FormatStyle::IGLS_OuterIndent) {
3465 if (OldLineLevel > 1 || (!Line->InPPDirective && OldLineLevel > 0))
3466 --Level;
3467 }
3468
3469 if (!IsGotoLabel && !Style.IndentCaseBlocks &&
3470 CommentsBeforeNextToken.empty() && FormatTok->is(Kind: tok::l_brace)) {
3471 CompoundStatementIndenter Indenter(this, Level,
3472 Style.BraceWrapping.AfterCaseLabel,
3473 Style.BraceWrapping.IndentBraces);
3474 parseBlock();
3475 if (FormatTok->is(Kind: tok::kw_break)) {
3476 if (Style.BraceWrapping.AfterControlStatement ==
3477 FormatStyle::BWACS_Always) {
3478 addUnwrappedLine();
3479 if (!Style.IndentCaseBlocks &&
3480 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) {
3481 ++Level;
3482 }
3483 }
3484 parseStructuralElement();
3485 }
3486 addUnwrappedLine();
3487 } else {
3488 if (FormatTok->is(Kind: tok::semi))
3489 nextToken();
3490 addUnwrappedLine();
3491 }
3492
3493 Level = OldLineLevel;
3494
3495 if (FormatTok->isNot(Kind: tok::l_brace)) {
3496 parseStructuralElement();
3497 addUnwrappedLine();
3498 }
3499}
3500
3501void UnwrappedLineParser::parseCaseLabel() {
3502 assert(FormatTok->is(tok::kw_case) && "'case' expected");
3503 auto *Case = FormatTok;
3504
3505 // FIXME: fix handling of complex expressions here.
3506 do {
3507 nextToken();
3508 if (FormatTok->is(Kind: tok::colon)) {
3509 FormatTok->setFinalizedType(TT_CaseLabelColon);
3510 break;
3511 }
3512 if (Style.isJava() && FormatTok->is(Kind: tok::arrow)) {
3513 FormatTok->setFinalizedType(TT_CaseLabelArrow);
3514 Case->setFinalizedType(TT_SwitchExpressionLabel);
3515 break;
3516 }
3517 } while (!eof());
3518 parseLabel();
3519}
3520
3521void UnwrappedLineParser::parseSwitch(bool IsExpr) {
3522 assert(FormatTok->is(tok::kw_switch) && "'switch' expected");
3523 nextToken();
3524 if (FormatTok->is(Kind: tok::l_paren))
3525 parseParens();
3526
3527 keepAncestorBraces();
3528
3529 if (FormatTok->is(Kind: tok::l_brace)) {
3530 CompoundStatementIndenter Indenter(this, Style, Line->Level);
3531 FormatTok->setFinalizedType(IsExpr ? TT_SwitchExpressionLBrace
3532 : TT_ControlStatementLBrace);
3533 if (IsExpr)
3534 parseChildBlock();
3535 else
3536 parseBlock();
3537 setPreviousRBraceType(TT_ControlStatementRBrace);
3538 if (!IsExpr)
3539 addUnwrappedLine();
3540 } else {
3541 addUnwrappedLine();
3542 ++Line->Level;
3543 parseStructuralElement();
3544 --Line->Level;
3545 }
3546
3547 if (Style.RemoveBracesLLVM)
3548 NestedTooDeep.pop_back();
3549}
3550
3551void UnwrappedLineParser::parseAccessSpecifier() {
3552 nextToken();
3553 // Understand Qt's slots.
3554 if (FormatTok->isOneOf(K1: Keywords.kw_slots, K2: Keywords.kw_qslots))
3555 nextToken();
3556 // Otherwise, we don't know what it is, and we'd better keep the next token.
3557 if (FormatTok->is(Kind: tok::colon))
3558 nextToken();
3559 addUnwrappedLine();
3560}
3561
3562/// Parses a requires, decides if it is a clause or an expression.
3563/// \pre The current token has to be the requires keyword.
3564/// \returns true if it parsed a clause.
3565bool UnwrappedLineParser::parseRequires(bool SeenEqual) {
3566 assert(FormatTok->is(tok::kw_requires) && "'requires' expected");
3567
3568 // We try to guess if it is a requires clause, or a requires expression. For
3569 // that we first check the next token.
3570 switch (Tokens->peekNextToken(/*SkipComment=*/true)->Tok.getKind()) {
3571 case tok::l_brace:
3572 // This can only be an expression, never a clause.
3573 parseRequiresExpression();
3574 return false;
3575 case tok::l_paren:
3576 // Clauses and expression can start with a paren, it's unclear what we have.
3577 break;
3578 default:
3579 // All other tokens can only be a clause.
3580 parseRequiresClause();
3581 return true;
3582 }
3583
3584 // Looking forward we would have to decide if there are function declaration
3585 // like arguments to the requires expression:
3586 // requires (T t) {
3587 // Or there is a constraint expression for the requires clause:
3588 // requires (C<T> && ...
3589
3590 // But first let's look behind.
3591 auto *PreviousNonComment = FormatTok->getPreviousNonComment();
3592
3593 if (!PreviousNonComment ||
3594 PreviousNonComment->is(TT: TT_RequiresExpressionLBrace)) {
3595 // If there is no token, or an expression left brace, we are a requires
3596 // clause within a requires expression.
3597 parseRequiresClause();
3598 return true;
3599 }
3600
3601 switch (PreviousNonComment->Tok.getKind()) {
3602 case tok::greater:
3603 case tok::r_paren:
3604 case tok::kw_noexcept:
3605 case tok::kw_const:
3606 case tok::star:
3607 case tok::amp:
3608 // This is a requires clause.
3609 parseRequiresClause();
3610 return true;
3611 case tok::ampamp: {
3612 // This can be either:
3613 // if (... && requires (T t) ...)
3614 // Or
3615 // void member(...) && requires (C<T> ...
3616 // We check the one token before that for a const:
3617 // void member(...) const && requires (C<T> ...
3618 auto PrevPrev = PreviousNonComment->getPreviousNonComment();
3619 if ((PrevPrev && PrevPrev->is(Kind: tok::kw_const)) || !SeenEqual) {
3620 parseRequiresClause();
3621 return true;
3622 }
3623 break;
3624 }
3625 default:
3626 if (PreviousNonComment->isTypeOrIdentifier(LangOpts)) {
3627 // This is a requires clause.
3628 parseRequiresClause();
3629 return true;
3630 }
3631 // It's an expression.
3632 parseRequiresExpression();
3633 return false;
3634 }
3635
3636 // Now we look forward and try to check if the paren content is a parameter
3637 // list. The parameters can be cv-qualified and contain references or
3638 // pointers.
3639 // So we want basically to check for TYPE NAME, but TYPE can contain all kinds
3640 // of stuff: typename, const, *, &, &&, ::, identifiers.
3641
3642 unsigned StoredPosition = Tokens->getPosition();
3643 FormatToken *NextToken = Tokens->getNextToken();
3644 int Lookahead = 0;
3645 auto PeekNext = [&Lookahead, &NextToken, this] {
3646 ++Lookahead;
3647 NextToken = Tokens->getNextToken();
3648 };
3649
3650 bool FoundType = false;
3651 bool LastWasColonColon = false;
3652 int OpenAngles = 0;
3653
3654 for (; Lookahead < 50; PeekNext()) {
3655 switch (NextToken->Tok.getKind()) {
3656 case tok::kw_volatile:
3657 case tok::kw_const:
3658 case tok::comma:
3659 if (OpenAngles == 0) {
3660 FormatTok = Tokens->setPosition(StoredPosition);
3661 parseRequiresExpression();
3662 return false;
3663 }
3664 break;
3665 case tok::eof:
3666 // Break out of the loop.
3667 Lookahead = 50;
3668 break;
3669 case tok::coloncolon:
3670 LastWasColonColon = true;
3671 break;
3672 case tok::kw_decltype:
3673 case tok::identifier:
3674 if (FoundType && !LastWasColonColon && OpenAngles == 0) {
3675 FormatTok = Tokens->setPosition(StoredPosition);
3676 parseRequiresExpression();
3677 return false;
3678 }
3679 FoundType = true;
3680 LastWasColonColon = false;
3681 break;
3682 case tok::less:
3683 ++OpenAngles;
3684 break;
3685 case tok::greater:
3686 --OpenAngles;
3687 break;
3688 default:
3689 if (NextToken->isTypeName(LangOpts)) {
3690 FormatTok = Tokens->setPosition(StoredPosition);
3691 parseRequiresExpression();
3692 return false;
3693 }
3694 break;
3695 }
3696 }
3697 // This seems to be a complicated expression, just assume it's a clause.
3698 FormatTok = Tokens->setPosition(StoredPosition);
3699 parseRequiresClause();
3700 return true;
3701}
3702
3703/// Parses a requires clause.
3704/// \sa parseRequiresExpression
3705///
3706/// Returns if it either has finished parsing the clause, or it detects, that
3707/// the clause is incorrect.
3708void UnwrappedLineParser::parseRequiresClause() {
3709 assert(FormatTok->is(tok::kw_requires) && "'requires' expected");
3710
3711 // If there is no previous token, we are within a requires expression,
3712 // otherwise we will always have the template or function declaration in front
3713 // of it.
3714 bool InRequiresExpression =
3715 !FormatTok->Previous ||
3716 FormatTok->Previous->is(TT: TT_RequiresExpressionLBrace);
3717
3718 FormatTok->setFinalizedType(InRequiresExpression
3719 ? TT_RequiresClauseInARequiresExpression
3720 : TT_RequiresClause);
3721 nextToken();
3722
3723 // NOTE: parseConstraintExpression is only ever called from this function.
3724 // It could be inlined into here.
3725 parseConstraintExpression();
3726
3727 if (!InRequiresExpression && FormatTok->Previous)
3728 FormatTok->Previous->ClosesRequiresClause = true;
3729}
3730
3731/// Parses a requires expression.
3732/// \sa parseRequiresClause
3733///
3734/// Returns if it either has finished parsing the expression, or it detects,
3735/// that the expression is incorrect.
3736void UnwrappedLineParser::parseRequiresExpression() {
3737 assert(FormatTok->is(tok::kw_requires) && "'requires' expected");
3738
3739 FormatTok->setFinalizedType(TT_RequiresExpression);
3740 nextToken();
3741
3742 if (FormatTok->is(Kind: tok::l_paren)) {
3743 FormatTok->setFinalizedType(TT_RequiresExpressionLParen);
3744 parseParens();
3745 }
3746
3747 if (FormatTok->is(Kind: tok::l_brace)) {
3748 FormatTok->setFinalizedType(TT_RequiresExpressionLBrace);
3749 parseChildBlock();
3750 }
3751}
3752
3753/// Parses a constraint expression.
3754///
3755/// This is the body of a requires clause. It returns, when the parsing is
3756/// complete, or the expression is incorrect.
3757void UnwrappedLineParser::parseConstraintExpression() {
3758 // The special handling for lambdas is needed since tryToParseLambda() eats a
3759 // token and if a requires expression is the last part of a requires clause
3760 // and followed by an attribute like [[nodiscard]] the ClosesRequiresClause is
3761 // not set on the correct token. Thus we need to be aware if we even expect a
3762 // lambda to be possible.
3763 // template <typename T> requires requires { ... } [[nodiscard]] ...;
3764 bool LambdaNextTimeAllowed = true;
3765
3766 // Within lambda declarations, it is permitted to put a requires clause after
3767 // its template parameter list, which would place the requires clause right
3768 // before the parentheses of the parameters of the lambda declaration. Thus,
3769 // we track if we expect to see grouping parentheses at all.
3770 // Without this check, `requires foo<T> (T t)` in the below example would be
3771 // seen as the whole requires clause, accidentally eating the parameters of
3772 // the lambda.
3773 // [&]<typename T> requires foo<T> (T t) { ... };
3774 bool TopLevelParensAllowed = true;
3775
3776 do {
3777 bool LambdaThisTimeAllowed = std::exchange(obj&: LambdaNextTimeAllowed, new_val: false);
3778
3779 switch (FormatTok->Tok.getKind()) {
3780 case tok::kw_requires:
3781 parseRequiresExpression();
3782 break;
3783
3784 case tok::l_paren:
3785 if (!TopLevelParensAllowed)
3786 return;
3787 parseParens(/*AmpAmpTokenType=*/StarAndAmpTokenType: TT_BinaryOperator);
3788 TopLevelParensAllowed = false;
3789 break;
3790
3791 case tok::l_square:
3792 if (!LambdaThisTimeAllowed || !tryToParseLambda())
3793 return;
3794 break;
3795
3796 case tok::kw_const:
3797 case tok::semi:
3798 case tok::kw_class:
3799 case tok::kw_struct:
3800 case tok::kw_union:
3801 return;
3802
3803 case tok::l_brace:
3804 // Potential function body.
3805 return;
3806
3807 case tok::ampamp:
3808 case tok::pipepipe:
3809 FormatTok->setFinalizedType(TT_BinaryOperator);
3810 nextToken();
3811 LambdaNextTimeAllowed = true;
3812 TopLevelParensAllowed = true;
3813 break;
3814
3815 case tok::comma:
3816 case tok::comment:
3817 LambdaNextTimeAllowed = LambdaThisTimeAllowed;
3818 nextToken();
3819 break;
3820
3821 case tok::kw_sizeof:
3822 case tok::greater:
3823 case tok::greaterequal:
3824 case tok::greatergreater:
3825 case tok::less:
3826 case tok::lessequal:
3827 case tok::lessless:
3828 case tok::equalequal:
3829 case tok::exclaim:
3830 case tok::exclaimequal:
3831 case tok::plus:
3832 case tok::minus:
3833 case tok::star:
3834 case tok::slash:
3835 LambdaNextTimeAllowed = true;
3836 TopLevelParensAllowed = true;
3837 // Just eat them.
3838 nextToken();
3839 break;
3840
3841 case tok::numeric_constant:
3842 case tok::coloncolon:
3843 case tok::kw_true:
3844 case tok::kw_false:
3845 TopLevelParensAllowed = false;
3846 // Just eat them.
3847 nextToken();
3848 break;
3849
3850 case tok::kw_static_cast:
3851 case tok::kw_const_cast:
3852 case tok::kw_reinterpret_cast:
3853 case tok::kw_dynamic_cast:
3854 nextToken();
3855 if (FormatTok->isNot(Kind: tok::less))
3856 return;
3857
3858 nextToken();
3859 parseBracedList(/*IsAngleBracket=*/true);
3860 break;
3861
3862 default:
3863 if (!FormatTok->Tok.getIdentifierInfo()) {
3864 // Identifiers are part of the default case, we check for more then
3865 // tok::identifier to handle builtin type traits.
3866 return;
3867 }
3868
3869 // We need to differentiate identifiers for a template deduction guide,
3870 // variables, or function return types (the constraint expression has
3871 // ended before that), and basically all other cases. But it's easier to
3872 // check the other way around.
3873 assert(FormatTok->Previous);
3874 switch (FormatTok->Previous->Tok.getKind()) {
3875 case tok::coloncolon: // Nested identifier.
3876 case tok::ampamp: // Start of a function or variable for the
3877 case tok::pipepipe: // constraint expression. (binary)
3878 case tok::exclaim: // The same as above, but unary.
3879 case tok::kw_requires: // Initial identifier of a requires clause.
3880 case tok::equal: // Initial identifier of a concept declaration.
3881 case tok::kw_template: // A dependent template.
3882 break;
3883 default:
3884 return;
3885 }
3886
3887 // Read identifier with optional template declaration.
3888 nextToken();
3889 if (FormatTok->is(Kind: tok::less)) {
3890 nextToken();
3891 parseBracedList(/*IsAngleBracket=*/true);
3892 }
3893 TopLevelParensAllowed = false;
3894 break;
3895 }
3896 } while (!eof());
3897}
3898
3899bool UnwrappedLineParser::parseEnum() {
3900 const FormatToken &InitialToken = *FormatTok;
3901
3902 // Won't be 'enum' for NS_ENUMs.
3903 if (FormatTok->is(Kind: tok::kw_enum))
3904 nextToken();
3905
3906 // In TypeScript, "enum" can also be used as property name, e.g. in interface
3907 // declarations. An "enum" keyword followed by a colon would be a syntax
3908 // error and thus assume it is just an identifier.
3909 if (Style.isJavaScript() && FormatTok->isOneOf(K1: tok::colon, K2: tok::question))
3910 return false;
3911
3912 // In protobuf, "enum" can be used as a field name.
3913 if (Style.Language == FormatStyle::LK_Proto && FormatTok->is(Kind: tok::equal))
3914 return false;
3915
3916 if (IsCpp) {
3917 // Eat up enum class ...
3918 if (FormatTok->isOneOf(K1: tok::kw_class, K2: tok::kw_struct))
3919 nextToken();
3920 while (FormatTok->is(Kind: tok::l_square))
3921 if (!handleCppAttributes())
3922 return false;
3923 }
3924
3925 while (FormatTok->Tok.getIdentifierInfo() ||
3926 FormatTok->isOneOf(K1: tok::colon, K2: tok::coloncolon, Ks: tok::less,
3927 Ks: tok::greater, Ks: tok::comma, Ks: tok::question,
3928 Ks: tok::l_square)) {
3929 if (FormatTok->is(Kind: tok::colon))
3930 FormatTok->setFinalizedType(TT_EnumUnderlyingTypeColon);
3931 if (Style.isVerilog()) {
3932 FormatTok->setFinalizedType(TT_VerilogDimensionedTypeName);
3933 nextToken();
3934 // In Verilog the base type can have dimensions.
3935 while (FormatTok->is(Kind: tok::l_square))
3936 parseSquare();
3937 } else {
3938 nextToken();
3939 }
3940 // We can have macros or attributes in between 'enum' and the enum name.
3941 if (FormatTok->is(Kind: tok::l_paren))
3942 parseParens();
3943 if (FormatTok->is(Kind: tok::identifier)) {
3944 nextToken();
3945 // If there are two identifiers in a row, this is likely an elaborate
3946 // return type. In Java, this can be "implements", etc.
3947 if (IsCpp && FormatTok->is(Kind: tok::identifier))
3948 return false;
3949 }
3950 }
3951
3952 // Just a declaration or something is wrong.
3953 if (FormatTok->isNot(Kind: tok::l_brace))
3954 return true;
3955 FormatTok->setFinalizedType(TT_EnumLBrace);
3956 FormatTok->setBlockKind(BK_Block);
3957
3958 if (Style.isJava()) {
3959 // Java enums are different.
3960 parseJavaEnumBody();
3961 return true;
3962 }
3963 if (Style.Language == FormatStyle::LK_Proto) {
3964 parseBlock(/*MustBeDeclaration=*/true);
3965 return true;
3966 }
3967
3968 const bool ManageWhitesmithsBraces =
3969 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths;
3970
3971 if (!Style.AllowShortEnumsOnASingleLine &&
3972 ShouldBreakBeforeBrace(Style, InitialToken,
3973 IsEmptyBlock: Tokens->peekNextToken()->is(Kind: tok::r_brace))) {
3974 addUnwrappedLine();
3975
3976 // If we're in Whitesmiths mode, indent the brace if we're not indenting
3977 // the whole block.
3978 if (ManageWhitesmithsBraces)
3979 ++Line->Level;
3980 }
3981 // Parse enum body.
3982 nextToken();
3983 if (!Style.AllowShortEnumsOnASingleLine) {
3984 addUnwrappedLine();
3985 if (!ManageWhitesmithsBraces)
3986 ++Line->Level;
3987 }
3988 const auto OpeningLineIndex = CurrentLines->empty()
3989 ? UnwrappedLine::kInvalidIndex
3990 : CurrentLines->size() - 1;
3991 bool HasError = !parseBracedList(/*IsAngleBracket=*/false, /*IsEnum=*/true);
3992 if (!Style.AllowShortEnumsOnASingleLine && !ManageWhitesmithsBraces)
3993 --Line->Level;
3994 if (HasError) {
3995 if (FormatTok->is(Kind: tok::semi))
3996 nextToken();
3997 addUnwrappedLine();
3998 }
3999 setPreviousRBraceType(TT_EnumRBrace);
4000 if (ManageWhitesmithsBraces)
4001 Line->MatchingOpeningBlockLineIndex = OpeningLineIndex;
4002 return true;
4003
4004 // There is no addUnwrappedLine() here so that we fall through to parsing a
4005 // structural element afterwards. Thus, in "enum A {} n, m;",
4006 // "} n, m;" will end up in one unwrapped line.
4007}
4008
4009bool UnwrappedLineParser::parseStructLike() {
4010 // parseRecord falls through and does not yet add an unwrapped line as a
4011 // record declaration or definition can start a structural element.
4012 parseRecord();
4013 // This does not apply to Java, JavaScript and C#.
4014 if (Style.isJava() || Style.isJavaScript() || Style.isCSharp()) {
4015 if (FormatTok->is(Kind: tok::semi))
4016 nextToken();
4017 addUnwrappedLine();
4018 return true;
4019 }
4020 return false;
4021}
4022
4023namespace {
4024// A class used to set and restore the Token position when peeking
4025// ahead in the token source.
4026class ScopedTokenPosition {
4027 unsigned StoredPosition;
4028 FormatTokenSource *Tokens;
4029
4030public:
4031 ScopedTokenPosition(FormatTokenSource *Tokens) : Tokens(Tokens) {
4032 assert(Tokens && "Tokens expected to not be null");
4033 StoredPosition = Tokens->getPosition();
4034 }
4035
4036 ~ScopedTokenPosition() { Tokens->setPosition(StoredPosition); }
4037};
4038} // namespace
4039
4040// Look to see if we have [[ by looking ahead, if
4041// its not then rewind to the original position.
4042bool UnwrappedLineParser::tryToParseSimpleAttribute() {
4043 ScopedTokenPosition AutoPosition(Tokens);
4044 FormatToken *Tok = Tokens->getNextToken();
4045 // We already read the first [ check for the second.
4046 if (Tok->isNot(Kind: tok::l_square))
4047 return false;
4048 // Double check that the attribute is just something
4049 // fairly simple.
4050 while (Tok->isNot(Kind: tok::eof)) {
4051 if (Tok->is(Kind: tok::r_square))
4052 break;
4053 Tok = Tokens->getNextToken();
4054 }
4055 if (Tok->is(Kind: tok::eof))
4056 return false;
4057 Tok = Tokens->getNextToken();
4058 if (Tok->isNot(Kind: tok::r_square))
4059 return false;
4060 Tok = Tokens->getNextToken();
4061 if (Tok->is(Kind: tok::semi))
4062 return false;
4063 return true;
4064}
4065
4066void UnwrappedLineParser::parseJavaEnumBody() {
4067 assert(FormatTok->is(tok::l_brace));
4068 const FormatToken *OpeningBrace = FormatTok;
4069
4070 // Determine whether the enum is simple, i.e. does not have a semicolon or
4071 // constants with class bodies. Simple enums can be formatted like braced
4072 // lists, contracted to a single line, etc.
4073 unsigned StoredPosition = Tokens->getPosition();
4074 bool IsSimple = true;
4075 FormatToken *Tok = Tokens->getNextToken();
4076 while (Tok->isNot(Kind: tok::eof)) {
4077 if (Tok->is(Kind: tok::r_brace))
4078 break;
4079 if (Tok->isOneOf(K1: tok::l_brace, K2: tok::semi)) {
4080 IsSimple = false;
4081 break;
4082 }
4083 // FIXME: This will also mark enums with braces in the arguments to enum
4084 // constants as "not simple". This is probably fine in practice, though.
4085 Tok = Tokens->getNextToken();
4086 }
4087 FormatTok = Tokens->setPosition(StoredPosition);
4088
4089 if (IsSimple) {
4090 nextToken();
4091 parseBracedList();
4092 addUnwrappedLine();
4093 return;
4094 }
4095
4096 // Parse the body of a more complex enum.
4097 // First add a line for everything up to the "{".
4098 nextToken();
4099 addUnwrappedLine();
4100 ++Line->Level;
4101
4102 // Parse the enum constants.
4103 while (!eof()) {
4104 if (FormatTok->is(Kind: tok::l_brace)) {
4105 // Parse the constant's class body.
4106 parseBlock(/*MustBeDeclaration=*/true, /*AddLevels=*/1u,
4107 /*MunchSemi=*/false);
4108 } else if (FormatTok->is(Kind: tok::l_paren)) {
4109 parseParens();
4110 } else if (FormatTok->is(Kind: tok::comma)) {
4111 nextToken();
4112 addUnwrappedLine();
4113 } else if (FormatTok->is(Kind: tok::semi)) {
4114 nextToken();
4115 addUnwrappedLine();
4116 break;
4117 } else if (FormatTok->is(Kind: tok::r_brace)) {
4118 addUnwrappedLine();
4119 break;
4120 } else {
4121 nextToken();
4122 }
4123 }
4124
4125 // Parse the class body after the enum's ";" if any.
4126 parseLevel(OpeningBrace);
4127 nextToken();
4128 --Line->Level;
4129 addUnwrappedLine();
4130}
4131
4132void UnwrappedLineParser::parseRecord(bool ParseAsExpr, bool IsJavaRecord) {
4133 assert(!IsJavaRecord || FormatTok->is(Keywords.kw_record));
4134 const FormatToken &InitialToken = *FormatTok;
4135 nextToken();
4136
4137 FormatToken *ClassName =
4138 IsJavaRecord && FormatTok->is(Kind: tok::identifier) ? FormatTok : nullptr;
4139 bool IsDerived = false;
4140 auto IsNonMacroIdentifier = [](const FormatToken *Tok) {
4141 return Tok->is(Kind: tok::identifier) && Tok->TokenText != Tok->TokenText.upper();
4142 };
4143 // JavaScript/TypeScript supports anonymous classes like:
4144 // a = class extends foo { }
4145 bool JSPastExtendsOrImplements = false;
4146 // The actual identifier can be a nested name specifier, and in macros
4147 // it is often token-pasted.
4148 // An [[attribute]] can be before the identifier.
4149 while (FormatTok->isOneOf(K1: tok::identifier, K2: tok::coloncolon, Ks: tok::hashhash,
4150 Ks: tok::kw_alignas, Ks: tok::l_square) ||
4151 FormatTok->isAttribute() ||
4152 ((Style.isJava() || Style.isJavaScript()) &&
4153 FormatTok->isOneOf(K1: tok::period, K2: tok::comma))) {
4154 if (Style.isJavaScript() &&
4155 FormatTok->isOneOf(K1: Keywords.kw_extends, K2: Keywords.kw_implements)) {
4156 JSPastExtendsOrImplements = true;
4157 // JavaScript/TypeScript supports inline object types in
4158 // extends/implements positions:
4159 // class Foo implements {bar: number} { }
4160 nextToken();
4161 if (FormatTok->is(Kind: tok::l_brace)) {
4162 tryToParseBracedList();
4163 continue;
4164 }
4165 }
4166 if (FormatTok->is(Kind: tok::l_square) && handleCppAttributes())
4167 continue;
4168 auto *Previous = FormatTok;
4169 nextToken();
4170 switch (FormatTok->Tok.getKind()) {
4171 case tok::l_paren:
4172 // We can have macros in between 'class' and the class name.
4173 if (IsJavaRecord || !IsNonMacroIdentifier(Previous) ||
4174 // e.g. `struct macro(a) S { int i; };`
4175 Previous->Previous == &InitialToken) {
4176 parseParens();
4177 }
4178 break;
4179 case tok::coloncolon:
4180 case tok::hashhash:
4181 break;
4182 default:
4183 if (JSPastExtendsOrImplements || ClassName ||
4184 Previous->isNot(Kind: tok::identifier) || Previous->is(TT: TT_AttributeMacro)) {
4185 break;
4186 }
4187 if (const auto Text = Previous->TokenText;
4188 Text.size() == 1 || Text != Text.upper()) {
4189 ClassName = Previous;
4190 }
4191 }
4192 }
4193
4194 auto IsListInitialization = [&] {
4195 if (!ClassName || IsDerived || JSPastExtendsOrImplements)
4196 return false;
4197 assert(FormatTok->is(tok::l_brace));
4198 const auto *Prev = FormatTok->getPreviousNonComment();
4199 assert(Prev);
4200 return Prev != ClassName && Prev->is(Kind: tok::identifier) &&
4201 Prev->isNot(Kind: Keywords.kw_final) && tryToParseBracedList();
4202 };
4203
4204 if (FormatTok->isOneOf(K1: tok::colon, K2: tok::less)) {
4205 int AngleNestingLevel = 0;
4206 do {
4207 if (FormatTok->is(Kind: tok::less))
4208 ++AngleNestingLevel;
4209 else if (FormatTok->is(Kind: tok::greater))
4210 --AngleNestingLevel;
4211
4212 if (AngleNestingLevel == 0) {
4213 if (FormatTok->is(Kind: tok::colon)) {
4214 IsDerived = true;
4215 } else if (!IsDerived && FormatTok->is(Kind: tok::identifier) &&
4216 FormatTok->Previous->is(Kind: tok::coloncolon)) {
4217 ClassName = FormatTok;
4218 } else if (FormatTok->is(Kind: tok::l_paren) &&
4219 IsNonMacroIdentifier(FormatTok->Previous)) {
4220 break;
4221 }
4222 }
4223 if (FormatTok->is(Kind: tok::l_brace)) {
4224 if (AngleNestingLevel == 0 && IsListInitialization())
4225 return;
4226 calculateBraceTypes(/*ExpectClassBody=*/true);
4227 if (!tryToParseBracedList())
4228 break;
4229 }
4230 if (FormatTok->is(Kind: tok::l_square)) {
4231 FormatToken *Previous = FormatTok->Previous;
4232 if (!Previous || (Previous->isNot(Kind: tok::r_paren) &&
4233 !Previous->isTypeOrIdentifier(LangOpts))) {
4234 // Don't try parsing a lambda if we had a closing parenthesis before,
4235 // it was probably a pointer to an array: int (*)[].
4236 if (!tryToParseLambda())
4237 continue;
4238 } else {
4239 parseSquare();
4240 continue;
4241 }
4242 }
4243 if (FormatTok->is(Kind: tok::semi))
4244 return;
4245 if (Style.isCSharp() && FormatTok->is(II: Keywords.kw_where)) {
4246 addUnwrappedLine();
4247 nextToken();
4248 parseCSharpGenericTypeConstraint();
4249 break;
4250 }
4251 nextToken();
4252 } while (!eof());
4253 }
4254
4255 auto GetBraceTypes =
4256 [](const FormatToken &RecordTok) -> std::pair<TokenType, TokenType> {
4257 switch (RecordTok.Tok.getKind()) {
4258 case tok::kw_class:
4259 return {TT_ClassLBrace, TT_ClassRBrace};
4260 case tok::kw_struct:
4261 return {TT_StructLBrace, TT_StructRBrace};
4262 case tok::kw_union:
4263 return {TT_UnionLBrace, TT_UnionRBrace};
4264 default:
4265 // Useful for e.g. interface.
4266 return {TT_RecordLBrace, TT_RecordRBrace};
4267 }
4268 };
4269 if (FormatTok->is(Kind: tok::l_brace)) {
4270 if (IsListInitialization())
4271 return;
4272 if (ClassName)
4273 ClassName->setFinalizedType(TT_ClassHeadName);
4274 auto [OpenBraceType, ClosingBraceType] = GetBraceTypes(InitialToken);
4275 FormatTok->setFinalizedType(OpenBraceType);
4276 if (ParseAsExpr) {
4277 parseChildBlock();
4278 } else {
4279 if (ShouldBreakBeforeBrace(Style, InitialToken,
4280 IsEmptyBlock: Tokens->peekNextToken()->is(Kind: tok::r_brace),
4281 IsJavaRecord)) {
4282 addUnwrappedLine();
4283 }
4284
4285 unsigned AddLevels = Style.IndentAccessModifiers ? 2u : 1u;
4286 parseBlock(/*MustBeDeclaration=*/true, AddLevels, /*MunchSemi=*/false);
4287 }
4288 setPreviousRBraceType(ClosingBraceType);
4289 }
4290 // There is no addUnwrappedLine() here so that we fall through to parsing a
4291 // structural element afterwards. Thus, in "class A {} n, m;",
4292 // "} n, m;" will end up in one unwrapped line.
4293}
4294
4295void UnwrappedLineParser::parseObjCMethod() {
4296 assert(FormatTok->isOneOf(tok::l_paren, tok::identifier) &&
4297 "'(' or identifier expected.");
4298 do {
4299 if (FormatTok->is(Kind: tok::semi)) {
4300 nextToken();
4301 addUnwrappedLine();
4302 return;
4303 } else if (FormatTok->is(Kind: tok::l_brace)) {
4304 if (Style.BraceWrapping.AfterFunction)
4305 addUnwrappedLine();
4306 parseBlock();
4307 addUnwrappedLine();
4308 return;
4309 } else {
4310 nextToken();
4311 }
4312 } while (!eof());
4313}
4314
4315void UnwrappedLineParser::parseObjCProtocolList() {
4316 assert(FormatTok->is(tok::less) && "'<' expected.");
4317 do {
4318 nextToken();
4319 // Early exit in case someone forgot a close angle.
4320 if (FormatTok->isOneOf(K1: tok::semi, K2: tok::l_brace, Ks: tok::objc_end))
4321 return;
4322 } while (!eof() && FormatTok->isNot(Kind: tok::greater));
4323 nextToken(); // Skip '>'.
4324}
4325
4326void UnwrappedLineParser::parseObjCUntilAtEnd() {
4327 do {
4328 if (FormatTok->is(Kind: tok::objc_end)) {
4329 nextToken();
4330 addUnwrappedLine();
4331 break;
4332 }
4333 if (FormatTok->is(Kind: tok::l_brace)) {
4334 parseBlock();
4335 // In ObjC interfaces, nothing should be following the "}".
4336 addUnwrappedLine();
4337 } else if (FormatTok->is(Kind: tok::r_brace)) {
4338 // Ignore stray "}". parseStructuralElement doesn't consume them.
4339 nextToken();
4340 addUnwrappedLine();
4341 } else if (FormatTok->isOneOf(K1: tok::minus, K2: tok::plus)) {
4342 nextToken();
4343 if (FormatTok->isOneOf(K1: tok::l_paren, K2: tok::identifier))
4344 parseObjCMethod();
4345 } else {
4346 parseStructuralElement();
4347 }
4348 } while (!eof());
4349}
4350
4351void UnwrappedLineParser::parseObjCInterfaceOrImplementation() {
4352 assert(FormatTok->isOneOf(tok::objc_interface, tok::objc_implementation));
4353 nextToken();
4354 nextToken(); // interface name
4355
4356 // @interface can be followed by a lightweight generic
4357 // specialization list, then either a base class or a category.
4358 if (FormatTok->is(Kind: tok::less))
4359 parseObjCLightweightGenerics();
4360 if (FormatTok->is(Kind: tok::colon)) {
4361 nextToken();
4362 nextToken(); // base class name
4363 // The base class can also have lightweight generics applied to it.
4364 if (FormatTok->is(Kind: tok::less))
4365 parseObjCLightweightGenerics();
4366 } else if (FormatTok->is(Kind: tok::l_paren)) {
4367 // Skip category, if present.
4368 parseParens();
4369 }
4370
4371 if (FormatTok->is(Kind: tok::less))
4372 parseObjCProtocolList();
4373
4374 if (FormatTok->is(Kind: tok::l_brace)) {
4375 if (Style.BraceWrapping.AfterObjCDeclaration)
4376 addUnwrappedLine();
4377 parseBlock(/*MustBeDeclaration=*/true);
4378 }
4379
4380 // With instance variables, this puts '}' on its own line. Without instance
4381 // variables, this ends the @interface line.
4382 addUnwrappedLine();
4383
4384 parseObjCUntilAtEnd();
4385}
4386
4387void UnwrappedLineParser::parseObjCLightweightGenerics() {
4388 assert(FormatTok->is(tok::less));
4389 // Unlike protocol lists, generic parameterizations support
4390 // nested angles:
4391 //
4392 // @interface Foo<ValueType : id <NSCopying, NSSecureCoding>> :
4393 // NSObject <NSCopying, NSSecureCoding>
4394 //
4395 // so we need to count how many open angles we have left.
4396 unsigned NumOpenAngles = 1;
4397 do {
4398 nextToken();
4399 // Early exit in case someone forgot a close angle.
4400 if (FormatTok->isOneOf(K1: tok::semi, K2: tok::l_brace, Ks: tok::objc_end))
4401 break;
4402 if (FormatTok->is(Kind: tok::less)) {
4403 ++NumOpenAngles;
4404 } else if (FormatTok->is(Kind: tok::greater)) {
4405 assert(NumOpenAngles > 0 && "'>' makes NumOpenAngles negative");
4406 --NumOpenAngles;
4407 }
4408 } while (!eof() && NumOpenAngles != 0);
4409 nextToken(); // Skip '>'.
4410}
4411
4412// Returns true for the declaration/definition form of @protocol,
4413// false for the expression form.
4414bool UnwrappedLineParser::parseObjCProtocol() {
4415 assert(FormatTok->is(tok::objc_protocol));
4416 nextToken();
4417
4418 if (FormatTok->is(Kind: tok::l_paren)) {
4419 // The expression form of @protocol, e.g. "Protocol* p = @protocol(foo);".
4420 return false;
4421 }
4422
4423 // The definition/declaration form,
4424 // @protocol Foo
4425 // - (int)someMethod;
4426 // @end
4427
4428 nextToken(); // protocol name
4429
4430 if (FormatTok->is(Kind: tok::less))
4431 parseObjCProtocolList();
4432
4433 // Check for protocol declaration.
4434 if (FormatTok->is(Kind: tok::semi)) {
4435 nextToken();
4436 addUnwrappedLine();
4437 return true;
4438 }
4439
4440 addUnwrappedLine();
4441 parseObjCUntilAtEnd();
4442 return true;
4443}
4444
4445void UnwrappedLineParser::parseJavaScriptEs6ImportExport() {
4446 bool IsImport = FormatTok->is(II: Keywords.kw_import);
4447 assert(IsImport || FormatTok->is(tok::kw_export));
4448 nextToken();
4449
4450 // Consume the "default" in "export default class/function".
4451 if (FormatTok->is(Kind: tok::kw_default))
4452 nextToken();
4453
4454 // Consume "async function", "function" and "default function", so that these
4455 // get parsed as free-standing JS functions, i.e. do not require a trailing
4456 // semicolon.
4457 if (FormatTok->is(II: Keywords.kw_async))
4458 nextToken();
4459 if (FormatTok->is(II: Keywords.kw_function)) {
4460 nextToken();
4461 return;
4462 }
4463
4464 // For imports, `export *`, `export {...}`, consume the rest of the line up
4465 // to the terminating `;`. For everything else, just return and continue
4466 // parsing the structural element, i.e. the declaration or expression for
4467 // `export default`.
4468 if (!IsImport && FormatTok->isNoneOf(Ks: tok::l_brace, Ks: tok::star) &&
4469 !FormatTok->isStringLiteral() &&
4470 !(FormatTok->is(II: Keywords.kw_type) &&
4471 Tokens->peekNextToken()->isOneOf(K1: tok::l_brace, K2: tok::star))) {
4472 return;
4473 }
4474
4475 while (!eof()) {
4476 if (FormatTok->is(Kind: tok::semi))
4477 return;
4478 if (Line->Tokens.empty()) {
4479 // Common issue: Automatic Semicolon Insertion wrapped the line, so the
4480 // import statement should terminate.
4481 return;
4482 }
4483 if (FormatTok->is(Kind: tok::l_brace)) {
4484 FormatTok->setBlockKind(BK_Block);
4485 nextToken();
4486 parseBracedList();
4487 } else {
4488 nextToken();
4489 }
4490 }
4491}
4492
4493void UnwrappedLineParser::parseStatementMacro() {
4494 nextToken();
4495 if (FormatTok->is(Kind: tok::l_paren))
4496 parseParens();
4497 if (FormatTok->is(Kind: tok::semi))
4498 nextToken();
4499 addUnwrappedLine();
4500}
4501
4502void UnwrappedLineParser::parseVerilogHierarchyIdentifier() {
4503 // consume things like a::`b.c[d:e] or a::*
4504 while (true) {
4505 if (FormatTok->isOneOf(K1: tok::star, K2: tok::period, Ks: tok::periodstar,
4506 Ks: tok::coloncolon, Ks: tok::hash) ||
4507 Keywords.isVerilogIdentifier(Tok: *FormatTok)) {
4508 nextToken();
4509 } else if (FormatTok->is(Kind: tok::l_square)) {
4510 parseSquare();
4511 } else {
4512 break;
4513 }
4514 }
4515}
4516
4517void UnwrappedLineParser::parseVerilogSensitivityList() {
4518 if (FormatTok->isNot(Kind: tok::at))
4519 return;
4520 nextToken();
4521 // A block event expression has 2 at signs.
4522 if (FormatTok->is(Kind: tok::at))
4523 nextToken();
4524 switch (FormatTok->Tok.getKind()) {
4525 case tok::star:
4526 nextToken();
4527 break;
4528 case tok::l_paren:
4529 parseParens();
4530 break;
4531 default:
4532 parseVerilogHierarchyIdentifier();
4533 break;
4534 }
4535}
4536
4537unsigned UnwrappedLineParser::parseVerilogHierarchyHeader() {
4538 unsigned AddLevels = 0;
4539
4540 if (FormatTok->is(II: Keywords.kw_clocking)) {
4541 nextToken();
4542 if (Keywords.isVerilogIdentifier(Tok: *FormatTok))
4543 nextToken();
4544 parseVerilogSensitivityList();
4545 if (FormatTok->is(Kind: tok::semi))
4546 nextToken();
4547 } else if (FormatTok->isOneOf(K1: tok::kw_case, K2: Keywords.kw_casex,
4548 Ks: Keywords.kw_casez, Ks: Keywords.kw_randcase,
4549 Ks: Keywords.kw_randsequence)) {
4550 if (Style.IndentCaseLabels)
4551 AddLevels++;
4552 nextToken();
4553 if (FormatTok->is(Kind: tok::l_paren)) {
4554 FormatTok->setFinalizedType(TT_ConditionLParen);
4555 parseParens();
4556 }
4557 if (FormatTok->isOneOf(K1: Keywords.kw_inside, K2: Keywords.kw_matches))
4558 nextToken();
4559 // The case header has no semicolon.
4560 } else {
4561 // "module" etc.
4562 nextToken();
4563 // all the words like the name of the module and specifiers like
4564 // "automatic" and the width of function return type
4565 while (true) {
4566 if (FormatTok->is(Kind: tok::l_square)) {
4567 auto Prev = FormatTok->getPreviousNonComment();
4568 if (Prev && Keywords.isVerilogIdentifier(Tok: *Prev))
4569 Prev->setFinalizedType(TT_VerilogDimensionedTypeName);
4570 parseSquare();
4571 } else if (Keywords.isVerilogIdentifier(Tok: *FormatTok) ||
4572 FormatTok->isOneOf(K1: tok::hash, K2: tok::hashhash, Ks: tok::coloncolon,
4573 Ks: Keywords.kw_automatic, Ks: tok::kw_static)) {
4574 nextToken();
4575 } else {
4576 break;
4577 }
4578 }
4579
4580 auto NewLine = [this]() {
4581 addUnwrappedLine();
4582 Line->IsContinuation = true;
4583 };
4584
4585 // package imports
4586 while (FormatTok->is(II: Keywords.kw_import)) {
4587 NewLine();
4588 nextToken();
4589 parseVerilogHierarchyIdentifier();
4590 if (FormatTok->is(Kind: tok::semi))
4591 nextToken();
4592 }
4593
4594 // parameters and ports
4595 if (FormatTok->is(II: Keywords.kw_verilogHash)) {
4596 NewLine();
4597 nextToken();
4598 if (FormatTok->is(Kind: tok::l_paren)) {
4599 FormatTok->setFinalizedType(TT_VerilogMultiLineListLParen);
4600 parseParens();
4601 }
4602 }
4603 if (FormatTok->is(Kind: tok::l_paren)) {
4604 NewLine();
4605 FormatTok->setFinalizedType(TT_VerilogMultiLineListLParen);
4606 parseParens();
4607 }
4608
4609 // extends and implements
4610 if (FormatTok->is(II: Keywords.kw_extends)) {
4611 NewLine();
4612 nextToken();
4613 parseVerilogHierarchyIdentifier();
4614 if (FormatTok->is(Kind: tok::l_paren))
4615 parseParens();
4616 }
4617 if (FormatTok->is(II: Keywords.kw_implements)) {
4618 NewLine();
4619 do {
4620 nextToken();
4621 parseVerilogHierarchyIdentifier();
4622 } while (FormatTok->is(Kind: tok::comma));
4623 }
4624
4625 // Coverage event for cover groups.
4626 if (FormatTok->is(Kind: tok::at)) {
4627 NewLine();
4628 parseVerilogSensitivityList();
4629 }
4630
4631 if (FormatTok->is(Kind: tok::semi))
4632 nextToken(/*LevelDifference=*/1);
4633 addUnwrappedLine();
4634 }
4635
4636 return AddLevels;
4637}
4638
4639void UnwrappedLineParser::parseVerilogTable() {
4640 assert(FormatTok->is(Keywords.kw_table));
4641 nextToken(/*LevelDifference=*/1);
4642 addUnwrappedLine();
4643
4644 auto InitialLevel = Line->Level++;
4645 while (!eof() && !Keywords.isVerilogEnd(Tok: *FormatTok)) {
4646 FormatToken *Tok = FormatTok;
4647 nextToken();
4648 if (Tok->is(Kind: tok::semi))
4649 addUnwrappedLine();
4650 else if (Tok->isOneOf(K1: tok::star, K2: tok::colon, Ks: tok::question, Ks: tok::minus))
4651 Tok->setFinalizedType(TT_VerilogTableItem);
4652 }
4653 Line->Level = InitialLevel;
4654 nextToken(/*LevelDifference=*/-1);
4655 addUnwrappedLine();
4656}
4657
4658void UnwrappedLineParser::parseVerilogCaseLabel() {
4659 // The label will get unindented in AnnotatingParser. If there are no leading
4660 // spaces, indent the rest here so that things inside the block will be
4661 // indented relative to things outside. We don't use parseLabel because we
4662 // don't know whether this colon is a label or a ternary expression at this
4663 // point.
4664 auto OrigLevel = Line->Level;
4665 auto FirstLine = CurrentLines->size();
4666 if (Line->Level == 0 || (Line->InPPDirective && Line->Level <= 1))
4667 ++Line->Level;
4668 else if (!Style.IndentCaseBlocks && Keywords.isVerilogBegin(Tok: *FormatTok))
4669 --Line->Level;
4670 parseStructuralElement();
4671 // Restore the indentation in both the new line and the line that has the
4672 // label.
4673 if (CurrentLines->size() > FirstLine)
4674 (*CurrentLines)[FirstLine].Level = OrigLevel;
4675 Line->Level = OrigLevel;
4676}
4677
4678void UnwrappedLineParser::parseVerilogExtern() {
4679 assert(
4680 FormatTok->isOneOf(tok::kw_extern, tok::kw_export, Keywords.kw_import));
4681 nextToken();
4682 // "DPI-C"
4683 if (FormatTok->is(Kind: tok::string_literal))
4684 nextToken();
4685 skipVerilogQualifiers();
4686 if (Keywords.isVerilogIdentifier(Tok: *FormatTok))
4687 nextToken();
4688 if (FormatTok->is(Kind: tok::equal))
4689 nextToken();
4690 if (Keywords.isVerilogHierarchy(Tok: *FormatTok))
4691 parseVerilogHierarchyHeader();
4692}
4693
4694void UnwrappedLineParser::skipVerilogQualifiers() {
4695 while (FormatTok->isOneOf(K1: tok::kw_protected, K2: tok::kw_virtual, Ks: tok::kw_static,
4696 Ks: Keywords.kw_rand, Ks: Keywords.kw_context,
4697 Ks: Keywords.kw_pure, Ks: Keywords.kw_randc,
4698 Ks: Keywords.kw_local)) {
4699 nextToken();
4700 }
4701}
4702
4703bool UnwrappedLineParser::containsExpansion(const UnwrappedLine &Line) const {
4704 for (const auto &N : Line.Tokens) {
4705 if (N.Tok->MacroCtx)
4706 return true;
4707 for (const UnwrappedLine &Child : N.Children)
4708 if (containsExpansion(Line: Child))
4709 return true;
4710 }
4711 return false;
4712}
4713
4714void UnwrappedLineParser::addUnwrappedLine(LineLevel AdjustLevel) {
4715 if (Line->Tokens.empty())
4716 return;
4717 LLVM_DEBUG({
4718 if (!parsingPPDirective()) {
4719 llvm::dbgs() << "Adding unwrapped line:\n";
4720 printDebugInfo(*Line);
4721 }
4722 });
4723
4724 // If this line closes a block when in Whitesmiths mode, remember that
4725 // information so that the level can be decreased after the line is added.
4726 // This has to happen after the addition of the line since the line itself
4727 // needs to be indented.
4728 bool ClosesWhitesmithsBlock =
4729 Line->MatchingOpeningBlockLineIndex != UnwrappedLine::kInvalidIndex &&
4730 Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths;
4731
4732 // If the current line was expanded from a macro call, we use it to
4733 // reconstruct an unwrapped line from the structure of the expanded unwrapped
4734 // line and the unexpanded token stream.
4735 if (!parsingPPDirective() && !InExpansion && containsExpansion(Line: *Line)) {
4736 if (!Reconstruct)
4737 Reconstruct.emplace(args&: Line->Level, args&: Unexpanded);
4738 Reconstruct->addLine(Line: *Line);
4739
4740 // While the reconstructed unexpanded lines are stored in the normal
4741 // flow of lines, the expanded lines are stored on the side to be analyzed
4742 // in an extra step.
4743 CurrentExpandedLines.push_back(Elt: std::move(*Line));
4744
4745 if (Reconstruct->finished()) {
4746 UnwrappedLine Reconstructed = std::move(*Reconstruct).takeResult();
4747 assert(!Reconstructed.Tokens.empty() &&
4748 "Reconstructed must at least contain the macro identifier.");
4749 assert(!parsingPPDirective());
4750 LLVM_DEBUG({
4751 llvm::dbgs() << "Adding unexpanded line:\n";
4752 printDebugInfo(Reconstructed);
4753 });
4754 ExpandedLines[Reconstructed.Tokens.begin()->Tok] = CurrentExpandedLines;
4755 Lines.push_back(Elt: std::move(Reconstructed));
4756 CurrentExpandedLines.clear();
4757 Reconstruct.reset();
4758 }
4759 } else {
4760 // At the top level we only get here when no unexpansion is going on, or
4761 // when conditional formatting led to unfinished macro reconstructions.
4762 assert(!Reconstruct || (CurrentLines != &Lines) || !PPStack.empty());
4763 CurrentLines->push_back(Elt: std::move(*Line));
4764 }
4765 Line->Tokens.clear();
4766 Line->MatchingOpeningBlockLineIndex = UnwrappedLine::kInvalidIndex;
4767 Line->FirstStartColumn = 0;
4768 Line->IsContinuation = false;
4769 Line->SeenDecltypeAuto = false;
4770 Line->IsModuleOrImportDecl = false;
4771
4772 if (ClosesWhitesmithsBlock && AdjustLevel == LineLevel::Remove)
4773 --Line->Level;
4774 if (!parsingPPDirective() && !PreprocessorDirectives.empty()) {
4775 CurrentLines->append(
4776 in_start: std::make_move_iterator(i: PreprocessorDirectives.begin()),
4777 in_end: std::make_move_iterator(i: PreprocessorDirectives.end()));
4778 PreprocessorDirectives.clear();
4779 }
4780 // Disconnect the current token from the last token on the previous line.
4781 FormatTok->Previous = nullptr;
4782}
4783
4784bool UnwrappedLineParser::eof() const { return FormatTok->is(Kind: tok::eof); }
4785
4786bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) {
4787 return (Line->InPPDirective || FormatTok.HasUnescapedNewline) &&
4788 FormatTok.NewlinesBefore > 0;
4789}
4790
4791// Checks if \p FormatTok is a line comment that continues the line comment
4792// section on \p Line.
4793static bool
4794continuesLineCommentSection(const FormatToken &FormatTok,
4795 const UnwrappedLine &Line, const FormatStyle &Style,
4796 const llvm::Regex &CommentPragmasRegex) {
4797 if (Line.Tokens.empty() || Style.ReflowComments != FormatStyle::RCS_Always)
4798 return false;
4799
4800 StringRef IndentContent = FormatTok.TokenText;
4801 if (FormatTok.TokenText.starts_with(Prefix: "//") ||
4802 FormatTok.TokenText.starts_with(Prefix: "/*")) {
4803 IndentContent = FormatTok.TokenText.substr(Start: 2);
4804 }
4805 if (CommentPragmasRegex.match(String: IndentContent))
4806 return false;
4807
4808 // If Line starts with a line comment, then FormatTok continues the comment
4809 // section if its original column is greater or equal to the original start
4810 // column of the line.
4811 //
4812 // Define the min column token of a line as follows: if a line ends in '{' or
4813 // contains a '{' followed by a line comment, then the min column token is
4814 // that '{'. Otherwise, the min column token of the line is the first token of
4815 // the line.
4816 //
4817 // If Line starts with a token other than a line comment, then FormatTok
4818 // continues the comment section if its original column is greater than the
4819 // original start column of the min column token of the line.
4820 //
4821 // For example, the second line comment continues the first in these cases:
4822 //
4823 // // first line
4824 // // second line
4825 //
4826 // and:
4827 //
4828 // // first line
4829 // // second line
4830 //
4831 // and:
4832 //
4833 // int i; // first line
4834 // // second line
4835 //
4836 // and:
4837 //
4838 // do { // first line
4839 // // second line
4840 // int i;
4841 // } while (true);
4842 //
4843 // and:
4844 //
4845 // enum {
4846 // a, // first line
4847 // // second line
4848 // b
4849 // };
4850 //
4851 // The second line comment doesn't continue the first in these cases:
4852 //
4853 // // first line
4854 // // second line
4855 //
4856 // and:
4857 //
4858 // int i; // first line
4859 // // second line
4860 //
4861 // and:
4862 //
4863 // do { // first line
4864 // // second line
4865 // int i;
4866 // } while (true);
4867 //
4868 // and:
4869 //
4870 // enum {
4871 // a, // first line
4872 // // second line
4873 // };
4874 const FormatToken *MinColumnToken = Line.Tokens.front().Tok;
4875
4876 // Scan for '{//'. If found, use the column of '{' as a min column for line
4877 // comment section continuation.
4878 const FormatToken *PreviousToken = nullptr;
4879 for (const UnwrappedLineNode &Node : Line.Tokens) {
4880 if (PreviousToken && PreviousToken->is(Kind: tok::l_brace) &&
4881 isLineComment(FormatTok: *Node.Tok)) {
4882 MinColumnToken = PreviousToken;
4883 break;
4884 }
4885 PreviousToken = Node.Tok;
4886
4887 // Grab the last newline preceding a token in this unwrapped line.
4888 if (Node.Tok->NewlinesBefore > 0)
4889 MinColumnToken = Node.Tok;
4890 }
4891 if (PreviousToken && PreviousToken->is(Kind: tok::l_brace))
4892 MinColumnToken = PreviousToken;
4893
4894 return continuesLineComment(FormatTok, /*Previous=*/Line.Tokens.back().Tok,
4895 MinColumnToken);
4896}
4897
4898void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) {
4899 bool JustComments = Line->Tokens.empty();
4900 for (FormatToken *Tok : CommentsBeforeNextToken) {
4901 // Line comments that belong to the same line comment section are put on the
4902 // same line since later we might want to reflow content between them.
4903 // Additional fine-grained breaking of line comment sections is controlled
4904 // by the class BreakableLineCommentSection in case it is desirable to keep
4905 // several line comment sections in the same unwrapped line.
4906 //
4907 // FIXME: Consider putting separate line comment sections as children to the
4908 // unwrapped line instead.
4909 Tok->ContinuesLineCommentSection =
4910 continuesLineCommentSection(FormatTok: *Tok, Line: *Line, Style, CommentPragmasRegex);
4911 if (isOnNewLine(FormatTok: *Tok) && JustComments && !Tok->ContinuesLineCommentSection)
4912 addUnwrappedLine();
4913 pushToken(Tok);
4914 }
4915 if (NewlineBeforeNext && JustComments)
4916 addUnwrappedLine();
4917 CommentsBeforeNextToken.clear();
4918}
4919
4920void UnwrappedLineParser::nextToken(int LevelDifference) {
4921 if (eof())
4922 return;
4923 flushComments(NewlineBeforeNext: isOnNewLine(FormatTok: *FormatTok));
4924 pushToken(Tok: FormatTok);
4925 FormatToken *Previous = FormatTok;
4926 if (!Style.isJavaScript())
4927 readToken(LevelDifference);
4928 else
4929 readTokenWithJavaScriptASI();
4930 FormatTok->Previous = Previous;
4931 if (Style.isVerilog()) {
4932 // Blocks in Verilog can have `begin` and `end` instead of braces. For
4933 // keywords like `begin`, we can't treat them the same as left braces
4934 // because some contexts require one of them. For example structs use
4935 // braces and if blocks use keywords, and a left brace can occur in an if
4936 // statement, but it is not a block. For keywords like `end`, we simply
4937 // treat them the same as right braces.
4938 if (Keywords.isVerilogEnd(Tok: *FormatTok))
4939 FormatTok->Tok.setKind(tok::r_brace);
4940 }
4941}
4942
4943void UnwrappedLineParser::distributeComments(
4944 const ArrayRef<FormatToken *> &Comments, const FormatToken *NextTok) {
4945 // Whether or not a line comment token continues a line is controlled by
4946 // the method continuesLineCommentSection, with the following caveat:
4947 //
4948 // Define a trail of Comments to be a nonempty proper postfix of Comments such
4949 // that each comment line from the trail is aligned with the next token, if
4950 // the next token exists. If a trail exists, the beginning of the maximal
4951 // trail is marked as a start of a new comment section.
4952 //
4953 // For example in this code:
4954 //
4955 // int a; // line about a
4956 // // line 1 about b
4957 // // line 2 about b
4958 // int b;
4959 //
4960 // the two lines about b form a maximal trail, so there are two sections, the
4961 // first one consisting of the single comment "// line about a" and the
4962 // second one consisting of the next two comments.
4963 if (Comments.empty())
4964 return;
4965 bool ShouldPushCommentsInCurrentLine = true;
4966 bool HasTrailAlignedWithNextToken = false;
4967 unsigned StartOfTrailAlignedWithNextToken = 0;
4968 if (NextTok) {
4969 // We are skipping the first element intentionally.
4970 for (unsigned i = Comments.size() - 1; i > 0; --i) {
4971 if (Comments[i]->OriginalColumn == NextTok->OriginalColumn) {
4972 HasTrailAlignedWithNextToken = true;
4973 StartOfTrailAlignedWithNextToken = i;
4974 }
4975 }
4976 }
4977 for (unsigned i = 0, e = Comments.size(); i < e; ++i) {
4978 FormatToken *FormatTok = Comments[i];
4979 if (HasTrailAlignedWithNextToken && i == StartOfTrailAlignedWithNextToken) {
4980 FormatTok->ContinuesLineCommentSection = false;
4981 } else {
4982 FormatTok->ContinuesLineCommentSection = continuesLineCommentSection(
4983 FormatTok: *FormatTok, Line: *Line, Style, CommentPragmasRegex);
4984 }
4985 if (!FormatTok->ContinuesLineCommentSection &&
4986 (isOnNewLine(FormatTok: *FormatTok) || FormatTok->IsFirst)) {
4987 ShouldPushCommentsInCurrentLine = false;
4988 }
4989 if (ShouldPushCommentsInCurrentLine)
4990 pushToken(Tok: FormatTok);
4991 else
4992 CommentsBeforeNextToken.push_back(Elt: FormatTok);
4993 }
4994}
4995
4996void UnwrappedLineParser::readToken(int LevelDifference) {
4997 SmallVector<FormatToken *, 1> Comments;
4998 bool PreviousWasComment = false;
4999 bool FirstNonCommentOnLine = false;
5000 do {
5001 FormatTok = Tokens->getNextToken();
5002 assert(FormatTok);
5003 while (FormatTok->isOneOf(K1: TT_ConflictStart, K2: TT_ConflictEnd,
5004 Ks: TT_ConflictAlternative)) {
5005 if (FormatTok->is(TT: TT_ConflictStart))
5006 conditionalCompilationStart(/*Unreachable=*/false);
5007 else if (FormatTok->is(TT: TT_ConflictAlternative))
5008 conditionalCompilationAlternative();
5009 else if (FormatTok->is(TT: TT_ConflictEnd))
5010 conditionalCompilationEnd();
5011 FormatTok = Tokens->getNextToken();
5012 FormatTok->MustBreakBefore = true;
5013 FormatTok->MustBreakBeforeFinalized = true;
5014 }
5015
5016 auto IsFirstNonCommentOnLine = [](bool FirstNonCommentOnLine,
5017 const FormatToken &Tok,
5018 bool PreviousWasComment) {
5019 auto IsFirstOnLine = [](const FormatToken &Tok) {
5020 return Tok.HasUnescapedNewline || Tok.IsFirst;
5021 };
5022
5023 // Consider preprocessor directives preceded by block comments as first
5024 // on line.
5025 if (PreviousWasComment)
5026 return FirstNonCommentOnLine || IsFirstOnLine(Tok);
5027 return IsFirstOnLine(Tok);
5028 };
5029
5030 FirstNonCommentOnLine = IsFirstNonCommentOnLine(
5031 FirstNonCommentOnLine, *FormatTok, PreviousWasComment);
5032 PreviousWasComment = FormatTok->is(Kind: tok::comment);
5033
5034 while (!Line->InPPDirective && FormatTok->is(Kind: tok::hash) &&
5035 FirstNonCommentOnLine) {
5036 // In Verilog, the backtick is used for macro invocations. In TableGen,
5037 // the single hash is used for the paste operator.
5038 const auto *Next = Tokens->peekNextToken();
5039 if ((Style.isVerilog() && !Keywords.isVerilogPPDirective(Tok: *Next)) ||
5040 (Style.isTableGen() &&
5041 Next->isNoneOf(Ks: tok::kw_else, Ks: tok::pp_define, Ks: tok::pp_ifdef,
5042 Ks: tok::pp_ifndef, Ks: tok::pp_endif))) {
5043 break;
5044 }
5045 distributeComments(Comments, NextTok: FormatTok);
5046 Comments.clear();
5047 // If there is an unfinished unwrapped line, we flush the preprocessor
5048 // directives only after that unwrapped line was finished later.
5049 bool SwitchToPreprocessorLines = !Line->Tokens.empty();
5050 ScopedLineState BlockState(*this, SwitchToPreprocessorLines);
5051 assert((LevelDifference >= 0 ||
5052 static_cast<unsigned>(-LevelDifference) <= Line->Level) &&
5053 "LevelDifference makes Line->Level negative");
5054 Line->Level += LevelDifference;
5055 // Comments stored before the preprocessor directive need to be output
5056 // before the preprocessor directive, at the same level as the
5057 // preprocessor directive, as we consider them to apply to the directive.
5058 if (Style.IndentPPDirectives == FormatStyle::PPDIS_BeforeHash &&
5059 PPBranchLevel > 0) {
5060 Line->Level += PPBranchLevel;
5061 }
5062 assert(Line->Level >= Line->UnbracedBodyLevel);
5063 Line->Level -= Line->UnbracedBodyLevel;
5064 flushComments(NewlineBeforeNext: isOnNewLine(FormatTok: *FormatTok));
5065 const bool IsEndIf = Tokens->peekNextToken()->is(Kind: tok::pp_endif);
5066 parsePPDirective();
5067 PreviousWasComment = FormatTok->is(Kind: tok::comment);
5068 FirstNonCommentOnLine = IsFirstNonCommentOnLine(
5069 FirstNonCommentOnLine, *FormatTok, PreviousWasComment);
5070 // If the #endif of a potential include guard is the last thing in the
5071 // file, then we found an include guard.
5072 if (IsEndIf && IncludeGuard == IG_Defined && PPBranchLevel == -1 &&
5073 getIncludeGuardState(Style: Style.IndentPPDirectives) == IG_Inited &&
5074 (eof() ||
5075 (PreviousWasComment &&
5076 Tokens->peekNextToken(/*SkipComment=*/true)->is(Kind: tok::eof)))) {
5077 IncludeGuard = IG_Found;
5078 }
5079 }
5080
5081 if (!PPStack.empty() && (PPStack.back().Kind == PP_Unreachable) &&
5082 !Line->InPPDirective) {
5083 continue;
5084 }
5085
5086 if (FormatTok->is(Kind: tok::identifier) &&
5087 Macros.defined(Name: FormatTok->TokenText) &&
5088 // FIXME: Allow expanding macros in preprocessor directives.
5089 !Line->InPPDirective) {
5090 FormatToken *ID = FormatTok;
5091 unsigned Position = Tokens->getPosition();
5092
5093 // To correctly parse the code, we need to replace the tokens of the macro
5094 // call with its expansion.
5095 auto PreCall = std::move(Line);
5096 Line.reset(p: new UnwrappedLine);
5097 bool OldInExpansion = InExpansion;
5098 InExpansion = true;
5099 // We parse the macro call into a new line.
5100 auto Args = parseMacroCall();
5101 InExpansion = OldInExpansion;
5102 assert(Line->Tokens.front().Tok == ID);
5103 // And remember the unexpanded macro call tokens.
5104 auto UnexpandedLine = std::move(Line);
5105 // Reset to the old line.
5106 Line = std::move(PreCall);
5107
5108 LLVM_DEBUG({
5109 llvm::dbgs() << "Macro call: " << ID->TokenText << "(";
5110 if (Args) {
5111 llvm::dbgs() << "(";
5112 for (const auto &Arg : Args.value())
5113 for (const auto &T : Arg)
5114 llvm::dbgs() << T->TokenText << " ";
5115 llvm::dbgs() << ")";
5116 }
5117 llvm::dbgs() << "\n";
5118 });
5119 if (Macros.objectLike(Name: ID->TokenText) && Args &&
5120 !Macros.hasArity(Name: ID->TokenText, Arity: Args->size())) {
5121 // The macro is either
5122 // - object-like, but we got argumnets, or
5123 // - overloaded to be both object-like and function-like, but none of
5124 // the function-like arities match the number of arguments.
5125 // Thus, expand as object-like macro.
5126 LLVM_DEBUG(llvm::dbgs()
5127 << "Macro \"" << ID->TokenText
5128 << "\" not overloaded for arity " << Args->size()
5129 << "or not function-like, using object-like overload.");
5130 Args.reset();
5131 UnexpandedLine->Tokens.resize(new_size: 1);
5132 Tokens->setPosition(Position);
5133 nextToken();
5134 assert(!Args && Macros.objectLike(ID->TokenText));
5135 }
5136 if ((!Args && Macros.objectLike(Name: ID->TokenText)) ||
5137 (Args && Macros.hasArity(Name: ID->TokenText, Arity: Args->size()))) {
5138 // Next, we insert the expanded tokens in the token stream at the
5139 // current position, and continue parsing.
5140 Unexpanded[ID] = std::move(UnexpandedLine);
5141 SmallVector<FormatToken *, 8> Expansion =
5142 Macros.expand(ID, OptionalArgs: std::move(Args));
5143 if (!Expansion.empty())
5144 FormatTok = Tokens->insertTokens(Tokens: Expansion);
5145
5146 LLVM_DEBUG({
5147 llvm::dbgs() << "Expanded: ";
5148 for (const auto &T : Expansion)
5149 llvm::dbgs() << T->TokenText << " ";
5150 llvm::dbgs() << "\n";
5151 });
5152 } else {
5153 LLVM_DEBUG({
5154 llvm::dbgs() << "Did not expand macro \"" << ID->TokenText
5155 << "\", because it was used ";
5156 if (Args)
5157 llvm::dbgs() << "with " << Args->size();
5158 else
5159 llvm::dbgs() << "without";
5160 llvm::dbgs() << " arguments, which doesn't match any definition.\n";
5161 });
5162 Tokens->setPosition(Position);
5163 FormatTok = ID;
5164 }
5165 }
5166
5167 if (FormatTok->isNot(Kind: tok::comment)) {
5168 distributeComments(Comments, NextTok: FormatTok);
5169 Comments.clear();
5170 return;
5171 }
5172
5173 Comments.push_back(Elt: FormatTok);
5174 } while (!eof());
5175
5176 distributeComments(Comments, NextTok: nullptr);
5177 Comments.clear();
5178}
5179
5180namespace {
5181template <typename Iterator>
5182void pushTokens(Iterator Begin, Iterator End,
5183 SmallVectorImpl<FormatToken *> &Into) {
5184 for (auto I = Begin; I != End; ++I) {
5185 Into.push_back(Elt: I->Tok);
5186 for (const auto &Child : I->Children)
5187 pushTokens(Child.Tokens.begin(), Child.Tokens.end(), Into);
5188 }
5189}
5190} // namespace
5191
5192std::optional<llvm::SmallVector<llvm::SmallVector<FormatToken *, 8>, 1>>
5193UnwrappedLineParser::parseMacroCall() {
5194 std::optional<llvm::SmallVector<llvm::SmallVector<FormatToken *, 8>, 1>> Args;
5195 assert(Line->Tokens.empty());
5196 nextToken();
5197 if (FormatTok->isNot(Kind: tok::l_paren))
5198 return Args;
5199 unsigned Position = Tokens->getPosition();
5200 FormatToken *Tok = FormatTok;
5201 nextToken();
5202 Args.emplace();
5203 auto ArgStart = std::prev(x: Line->Tokens.end());
5204
5205 int Parens = 0;
5206 do {
5207 switch (FormatTok->Tok.getKind()) {
5208 case tok::l_paren:
5209 ++Parens;
5210 nextToken();
5211 break;
5212 case tok::r_paren: {
5213 if (Parens > 0) {
5214 --Parens;
5215 nextToken();
5216 break;
5217 }
5218 Args->push_back(Elt: {});
5219 pushTokens(Begin: std::next(x: ArgStart), End: Line->Tokens.end(), Into&: Args->back());
5220 nextToken();
5221 return Args;
5222 }
5223 case tok::comma: {
5224 if (Parens > 0) {
5225 nextToken();
5226 break;
5227 }
5228 Args->push_back(Elt: {});
5229 pushTokens(Begin: std::next(x: ArgStart), End: Line->Tokens.end(), Into&: Args->back());
5230 nextToken();
5231 ArgStart = std::prev(x: Line->Tokens.end());
5232 break;
5233 }
5234 default:
5235 nextToken();
5236 break;
5237 }
5238 } while (!eof());
5239 Line->Tokens.resize(new_size: 1);
5240 Tokens->setPosition(Position);
5241 FormatTok = Tok;
5242 return {};
5243}
5244
5245void UnwrappedLineParser::pushToken(FormatToken *Tok) {
5246 Line->Tokens.push_back(x: UnwrappedLineNode(Tok));
5247 if (AtEndOfPPLine) {
5248 auto &Tok = *Line->Tokens.back().Tok;
5249 Tok.MustBreakBefore = true;
5250 Tok.MustBreakBeforeFinalized = true;
5251 Tok.FirstAfterPPLine = true;
5252 AtEndOfPPLine = false;
5253 }
5254}
5255
5256} // end namespace format
5257} // end namespace clang
5258