xref: /freebsd/contrib/llvm-project/clang/lib/Format/BreakableToken.cpp (revision 162ae9c834f6d9f9cb443bd62cceb23e0b5fef48)
1 //===--- BreakableToken.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 /// Contains implementation of BreakableToken class and classes derived
11 /// from it.
12 ///
13 //===----------------------------------------------------------------------===//
14 
15 #include "BreakableToken.h"
16 #include "ContinuationIndenter.h"
17 #include "clang/Basic/CharInfo.h"
18 #include "clang/Format/Format.h"
19 #include "llvm/ADT/STLExtras.h"
20 #include "llvm/Support/Debug.h"
21 #include <algorithm>
22 
23 #define DEBUG_TYPE "format-token-breaker"
24 
25 namespace clang {
26 namespace format {
27 
28 static const char *const Blanks = " \t\v\f\r";
29 static bool IsBlank(char C) {
30   switch (C) {
31   case ' ':
32   case '\t':
33   case '\v':
34   case '\f':
35   case '\r':
36     return true;
37   default:
38     return false;
39   }
40 }
41 
42 static StringRef getLineCommentIndentPrefix(StringRef Comment,
43                                             const FormatStyle &Style) {
44   static const char *const KnownCStylePrefixes[] = {"///<", "//!<", "///", "//",
45                                                     "//!"};
46   static const char *const KnownTextProtoPrefixes[] = {"//", "#", "##", "###",
47                                                        "####"};
48   ArrayRef<const char *> KnownPrefixes(KnownCStylePrefixes);
49   if (Style.Language == FormatStyle::LK_TextProto)
50     KnownPrefixes = KnownTextProtoPrefixes;
51 
52   StringRef LongestPrefix;
53   for (StringRef KnownPrefix : KnownPrefixes) {
54     if (Comment.startswith(KnownPrefix)) {
55       size_t PrefixLength = KnownPrefix.size();
56       while (PrefixLength < Comment.size() && Comment[PrefixLength] == ' ')
57         ++PrefixLength;
58       if (PrefixLength > LongestPrefix.size())
59         LongestPrefix = Comment.substr(0, PrefixLength);
60     }
61   }
62   return LongestPrefix;
63 }
64 
65 static BreakableToken::Split
66 getCommentSplit(StringRef Text, unsigned ContentStartColumn,
67                 unsigned ColumnLimit, unsigned TabWidth,
68                 encoding::Encoding Encoding, const FormatStyle &Style,
69                 bool DecorationEndsWithStar = false) {
70   LLVM_DEBUG(llvm::dbgs() << "Comment split: \"" << Text
71                           << "\", Column limit: " << ColumnLimit
72                           << ", Content start: " << ContentStartColumn << "\n");
73   if (ColumnLimit <= ContentStartColumn + 1)
74     return BreakableToken::Split(StringRef::npos, 0);
75 
76   unsigned MaxSplit = ColumnLimit - ContentStartColumn + 1;
77   unsigned MaxSplitBytes = 0;
78 
79   for (unsigned NumChars = 0;
80        NumChars < MaxSplit && MaxSplitBytes < Text.size();) {
81     unsigned BytesInChar =
82         encoding::getCodePointNumBytes(Text[MaxSplitBytes], Encoding);
83     NumChars +=
84         encoding::columnWidthWithTabs(Text.substr(MaxSplitBytes, BytesInChar),
85                                       ContentStartColumn, TabWidth, Encoding);
86     MaxSplitBytes += BytesInChar;
87   }
88 
89   StringRef::size_type SpaceOffset = Text.find_last_of(Blanks, MaxSplitBytes);
90 
91   static auto *const kNumberedListRegexp = new llvm::Regex("^[1-9][0-9]?\\.");
92   while (SpaceOffset != StringRef::npos) {
93     // Do not split before a number followed by a dot: this would be interpreted
94     // as a numbered list, which would prevent re-flowing in subsequent passes.
95     if (kNumberedListRegexp->match(Text.substr(SpaceOffset).ltrim(Blanks)))
96       SpaceOffset = Text.find_last_of(Blanks, SpaceOffset);
97     // In JavaScript, some @tags can be followed by {, and machinery that parses
98     // these comments will fail to understand the comment if followed by a line
99     // break. So avoid ever breaking before a {.
100     else if (Style.Language == FormatStyle::LK_JavaScript &&
101              SpaceOffset + 1 < Text.size() && Text[SpaceOffset + 1] == '{')
102       SpaceOffset = Text.find_last_of(Blanks, SpaceOffset);
103     else
104       break;
105   }
106 
107   if (SpaceOffset == StringRef::npos ||
108       // Don't break at leading whitespace.
109       Text.find_last_not_of(Blanks, SpaceOffset) == StringRef::npos) {
110     // Make sure that we don't break at leading whitespace that
111     // reaches past MaxSplit.
112     StringRef::size_type FirstNonWhitespace = Text.find_first_not_of(Blanks);
113     if (FirstNonWhitespace == StringRef::npos)
114       // If the comment is only whitespace, we cannot split.
115       return BreakableToken::Split(StringRef::npos, 0);
116     SpaceOffset = Text.find_first_of(
117         Blanks, std::max<unsigned>(MaxSplitBytes, FirstNonWhitespace));
118   }
119   if (SpaceOffset != StringRef::npos && SpaceOffset != 0) {
120     // adaptStartOfLine will break after lines starting with /** if the comment
121     // is broken anywhere. Avoid emitting this break twice here.
122     // Example: in /** longtextcomesherethatbreaks */ (with ColumnLimit 20) will
123     // insert a break after /**, so this code must not insert the same break.
124     if (SpaceOffset == 1 && Text[SpaceOffset - 1] == '*')
125       return BreakableToken::Split(StringRef::npos, 0);
126     StringRef BeforeCut = Text.substr(0, SpaceOffset).rtrim(Blanks);
127     StringRef AfterCut = Text.substr(SpaceOffset);
128     // Don't trim the leading blanks if it would create a */ after the break.
129     if (!DecorationEndsWithStar || AfterCut.size() <= 1 || AfterCut[1] != '/')
130       AfterCut = AfterCut.ltrim(Blanks);
131     return BreakableToken::Split(BeforeCut.size(),
132                                  AfterCut.begin() - BeforeCut.end());
133   }
134   return BreakableToken::Split(StringRef::npos, 0);
135 }
136 
137 static BreakableToken::Split
138 getStringSplit(StringRef Text, unsigned UsedColumns, unsigned ColumnLimit,
139                unsigned TabWidth, encoding::Encoding Encoding) {
140   // FIXME: Reduce unit test case.
141   if (Text.empty())
142     return BreakableToken::Split(StringRef::npos, 0);
143   if (ColumnLimit <= UsedColumns)
144     return BreakableToken::Split(StringRef::npos, 0);
145   unsigned MaxSplit = ColumnLimit - UsedColumns;
146   StringRef::size_type SpaceOffset = 0;
147   StringRef::size_type SlashOffset = 0;
148   StringRef::size_type WordStartOffset = 0;
149   StringRef::size_type SplitPoint = 0;
150   for (unsigned Chars = 0;;) {
151     unsigned Advance;
152     if (Text[0] == '\\') {
153       Advance = encoding::getEscapeSequenceLength(Text);
154       Chars += Advance;
155     } else {
156       Advance = encoding::getCodePointNumBytes(Text[0], Encoding);
157       Chars += encoding::columnWidthWithTabs(
158           Text.substr(0, Advance), UsedColumns + Chars, TabWidth, Encoding);
159     }
160 
161     if (Chars > MaxSplit || Text.size() <= Advance)
162       break;
163 
164     if (IsBlank(Text[0]))
165       SpaceOffset = SplitPoint;
166     if (Text[0] == '/')
167       SlashOffset = SplitPoint;
168     if (Advance == 1 && !isAlphanumeric(Text[0]))
169       WordStartOffset = SplitPoint;
170 
171     SplitPoint += Advance;
172     Text = Text.substr(Advance);
173   }
174 
175   if (SpaceOffset != 0)
176     return BreakableToken::Split(SpaceOffset + 1, 0);
177   if (SlashOffset != 0)
178     return BreakableToken::Split(SlashOffset + 1, 0);
179   if (WordStartOffset != 0)
180     return BreakableToken::Split(WordStartOffset + 1, 0);
181   if (SplitPoint != 0)
182     return BreakableToken::Split(SplitPoint, 0);
183   return BreakableToken::Split(StringRef::npos, 0);
184 }
185 
186 bool switchesFormatting(const FormatToken &Token) {
187   assert((Token.is(TT_BlockComment) || Token.is(TT_LineComment)) &&
188          "formatting regions are switched by comment tokens");
189   StringRef Content = Token.TokenText.substr(2).ltrim();
190   return Content.startswith("clang-format on") ||
191          Content.startswith("clang-format off");
192 }
193 
194 unsigned
195 BreakableToken::getLengthAfterCompression(unsigned RemainingTokenColumns,
196                                           Split Split) const {
197   // Example: consider the content
198   // lala  lala
199   // - RemainingTokenColumns is the original number of columns, 10;
200   // - Split is (4, 2), denoting the two spaces between the two words;
201   //
202   // We compute the number of columns when the split is compressed into a single
203   // space, like:
204   // lala lala
205   //
206   // FIXME: Correctly measure the length of whitespace in Split.second so it
207   // works with tabs.
208   return RemainingTokenColumns + 1 - Split.second;
209 }
210 
211 unsigned BreakableStringLiteral::getLineCount() const { return 1; }
212 
213 unsigned BreakableStringLiteral::getRangeLength(unsigned LineIndex,
214                                                 unsigned Offset,
215                                                 StringRef::size_type Length,
216                                                 unsigned StartColumn) const {
217   llvm_unreachable("Getting the length of a part of the string literal "
218                    "indicates that the code tries to reflow it.");
219 }
220 
221 unsigned
222 BreakableStringLiteral::getRemainingLength(unsigned LineIndex, unsigned Offset,
223                                            unsigned StartColumn) const {
224   return UnbreakableTailLength + Postfix.size() +
225          encoding::columnWidthWithTabs(Line.substr(Offset, StringRef::npos),
226                                        StartColumn, Style.TabWidth, Encoding);
227 }
228 
229 unsigned BreakableStringLiteral::getContentStartColumn(unsigned LineIndex,
230                                                        bool Break) const {
231   return StartColumn + Prefix.size();
232 }
233 
234 BreakableStringLiteral::BreakableStringLiteral(
235     const FormatToken &Tok, unsigned StartColumn, StringRef Prefix,
236     StringRef Postfix, unsigned UnbreakableTailLength, bool InPPDirective,
237     encoding::Encoding Encoding, const FormatStyle &Style)
238     : BreakableToken(Tok, InPPDirective, Encoding, Style),
239       StartColumn(StartColumn), Prefix(Prefix), Postfix(Postfix),
240       UnbreakableTailLength(UnbreakableTailLength) {
241   assert(Tok.TokenText.startswith(Prefix) && Tok.TokenText.endswith(Postfix));
242   Line = Tok.TokenText.substr(
243       Prefix.size(), Tok.TokenText.size() - Prefix.size() - Postfix.size());
244 }
245 
246 BreakableToken::Split BreakableStringLiteral::getSplit(
247     unsigned LineIndex, unsigned TailOffset, unsigned ColumnLimit,
248     unsigned ContentStartColumn, llvm::Regex &CommentPragmasRegex) const {
249   return getStringSplit(Line.substr(TailOffset), ContentStartColumn,
250                         ColumnLimit - Postfix.size(), Style.TabWidth, Encoding);
251 }
252 
253 void BreakableStringLiteral::insertBreak(unsigned LineIndex,
254                                          unsigned TailOffset, Split Split,
255                                          unsigned ContentIndent,
256                                          WhitespaceManager &Whitespaces) const {
257   Whitespaces.replaceWhitespaceInToken(
258       Tok, Prefix.size() + TailOffset + Split.first, Split.second, Postfix,
259       Prefix, InPPDirective, 1, StartColumn);
260 }
261 
262 BreakableComment::BreakableComment(const FormatToken &Token,
263                                    unsigned StartColumn, bool InPPDirective,
264                                    encoding::Encoding Encoding,
265                                    const FormatStyle &Style)
266     : BreakableToken(Token, InPPDirective, Encoding, Style),
267       StartColumn(StartColumn) {}
268 
269 unsigned BreakableComment::getLineCount() const { return Lines.size(); }
270 
271 BreakableToken::Split
272 BreakableComment::getSplit(unsigned LineIndex, unsigned TailOffset,
273                            unsigned ColumnLimit, unsigned ContentStartColumn,
274                            llvm::Regex &CommentPragmasRegex) const {
275   // Don't break lines matching the comment pragmas regex.
276   if (CommentPragmasRegex.match(Content[LineIndex]))
277     return Split(StringRef::npos, 0);
278   return getCommentSplit(Content[LineIndex].substr(TailOffset),
279                          ContentStartColumn, ColumnLimit, Style.TabWidth,
280                          Encoding, Style);
281 }
282 
283 void BreakableComment::compressWhitespace(
284     unsigned LineIndex, unsigned TailOffset, Split Split,
285     WhitespaceManager &Whitespaces) const {
286   StringRef Text = Content[LineIndex].substr(TailOffset);
287   // Text is relative to the content line, but Whitespaces operates relative to
288   // the start of the corresponding token, so compute the start of the Split
289   // that needs to be compressed into a single space relative to the start of
290   // its token.
291   unsigned BreakOffsetInToken =
292       Text.data() - tokenAt(LineIndex).TokenText.data() + Split.first;
293   unsigned CharsToRemove = Split.second;
294   Whitespaces.replaceWhitespaceInToken(
295       tokenAt(LineIndex), BreakOffsetInToken, CharsToRemove, "", "",
296       /*InPPDirective=*/false, /*Newlines=*/0, /*Spaces=*/1);
297 }
298 
299 const FormatToken &BreakableComment::tokenAt(unsigned LineIndex) const {
300   return Tokens[LineIndex] ? *Tokens[LineIndex] : Tok;
301 }
302 
303 static bool mayReflowContent(StringRef Content) {
304   Content = Content.trim(Blanks);
305   // Lines starting with '@' commonly have special meaning.
306   // Lines starting with '-', '-#', '+' or '*' are bulleted/numbered lists.
307   bool hasSpecialMeaningPrefix = false;
308   for (StringRef Prefix :
309        {"@", "TODO", "FIXME", "XXX", "-# ", "- ", "+ ", "* "}) {
310     if (Content.startswith(Prefix)) {
311       hasSpecialMeaningPrefix = true;
312       break;
313     }
314   }
315 
316   // Numbered lists may also start with a number followed by '.'
317   // To avoid issues if a line starts with a number which is actually the end
318   // of a previous line, we only consider numbers with up to 2 digits.
319   static auto *const kNumberedListRegexp = new llvm::Regex("^[1-9][0-9]?\\. ");
320   hasSpecialMeaningPrefix =
321       hasSpecialMeaningPrefix || kNumberedListRegexp->match(Content);
322 
323   // Simple heuristic for what to reflow: content should contain at least two
324   // characters and either the first or second character must be
325   // non-punctuation.
326   return Content.size() >= 2 && !hasSpecialMeaningPrefix &&
327          !Content.endswith("\\") &&
328          // Note that this is UTF-8 safe, since if isPunctuation(Content[0]) is
329          // true, then the first code point must be 1 byte long.
330          (!isPunctuation(Content[0]) || !isPunctuation(Content[1]));
331 }
332 
333 BreakableBlockComment::BreakableBlockComment(
334     const FormatToken &Token, unsigned StartColumn,
335     unsigned OriginalStartColumn, bool FirstInLine, bool InPPDirective,
336     encoding::Encoding Encoding, const FormatStyle &Style, bool UseCRLF)
337     : BreakableComment(Token, StartColumn, InPPDirective, Encoding, Style),
338       DelimitersOnNewline(false),
339       UnbreakableTailLength(Token.UnbreakableTailLength) {
340   assert(Tok.is(TT_BlockComment) &&
341          "block comment section must start with a block comment");
342 
343   StringRef TokenText(Tok.TokenText);
344   assert(TokenText.startswith("/*") && TokenText.endswith("*/"));
345   TokenText.substr(2, TokenText.size() - 4).split(Lines,
346                                                   UseCRLF ? "\r\n" : "\n");
347 
348   int IndentDelta = StartColumn - OriginalStartColumn;
349   Content.resize(Lines.size());
350   Content[0] = Lines[0];
351   ContentColumn.resize(Lines.size());
352   // Account for the initial '/*'.
353   ContentColumn[0] = StartColumn + 2;
354   Tokens.resize(Lines.size());
355   for (size_t i = 1; i < Lines.size(); ++i)
356     adjustWhitespace(i, IndentDelta);
357 
358   // Align decorations with the column of the star on the first line,
359   // that is one column after the start "/*".
360   DecorationColumn = StartColumn + 1;
361 
362   // Account for comment decoration patterns like this:
363   //
364   // /*
365   // ** blah blah blah
366   // */
367   if (Lines.size() >= 2 && Content[1].startswith("**") &&
368       static_cast<unsigned>(ContentColumn[1]) == StartColumn) {
369     DecorationColumn = StartColumn;
370   }
371 
372   Decoration = "* ";
373   if (Lines.size() == 1 && !FirstInLine) {
374     // Comments for which FirstInLine is false can start on arbitrary column,
375     // and available horizontal space can be too small to align consecutive
376     // lines with the first one.
377     // FIXME: We could, probably, align them to current indentation level, but
378     // now we just wrap them without stars.
379     Decoration = "";
380   }
381   for (size_t i = 1, e = Lines.size(); i < e && !Decoration.empty(); ++i) {
382     // If the last line is empty, the closing "*/" will have a star.
383     if (i + 1 == e && Content[i].empty())
384       break;
385     if (!Content[i].empty() && i + 1 != e && Decoration.startswith(Content[i]))
386       continue;
387     while (!Content[i].startswith(Decoration))
388       Decoration = Decoration.substr(0, Decoration.size() - 1);
389   }
390 
391   LastLineNeedsDecoration = true;
392   IndentAtLineBreak = ContentColumn[0] + 1;
393   for (size_t i = 1, e = Lines.size(); i < e; ++i) {
394     if (Content[i].empty()) {
395       if (i + 1 == e) {
396         // Empty last line means that we already have a star as a part of the
397         // trailing */. We also need to preserve whitespace, so that */ is
398         // correctly indented.
399         LastLineNeedsDecoration = false;
400         // Align the star in the last '*/' with the stars on the previous lines.
401         if (e >= 2 && !Decoration.empty()) {
402           ContentColumn[i] = DecorationColumn;
403         }
404       } else if (Decoration.empty()) {
405         // For all other lines, set the start column to 0 if they're empty, so
406         // we do not insert trailing whitespace anywhere.
407         ContentColumn[i] = 0;
408       }
409       continue;
410     }
411 
412     // The first line already excludes the star.
413     // The last line excludes the star if LastLineNeedsDecoration is false.
414     // For all other lines, adjust the line to exclude the star and
415     // (optionally) the first whitespace.
416     unsigned DecorationSize = Decoration.startswith(Content[i])
417                                   ? Content[i].size()
418                                   : Decoration.size();
419     if (DecorationSize) {
420       ContentColumn[i] = DecorationColumn + DecorationSize;
421     }
422     Content[i] = Content[i].substr(DecorationSize);
423     if (!Decoration.startswith(Content[i]))
424       IndentAtLineBreak =
425           std::min<int>(IndentAtLineBreak, std::max(0, ContentColumn[i]));
426   }
427   IndentAtLineBreak = std::max<unsigned>(IndentAtLineBreak, Decoration.size());
428 
429   // Detect a multiline jsdoc comment and set DelimitersOnNewline in that case.
430   if (Style.Language == FormatStyle::LK_JavaScript ||
431       Style.Language == FormatStyle::LK_Java) {
432     if ((Lines[0] == "*" || Lines[0].startswith("* ")) && Lines.size() > 1) {
433       // This is a multiline jsdoc comment.
434       DelimitersOnNewline = true;
435     } else if (Lines[0].startswith("* ") && Lines.size() == 1) {
436       // Detect a long single-line comment, like:
437       // /** long long long */
438       // Below, '2' is the width of '*/'.
439       unsigned EndColumn =
440           ContentColumn[0] +
441           encoding::columnWidthWithTabs(Lines[0], ContentColumn[0],
442                                         Style.TabWidth, Encoding) +
443           2;
444       DelimitersOnNewline = EndColumn > Style.ColumnLimit;
445     }
446   }
447 
448   LLVM_DEBUG({
449     llvm::dbgs() << "IndentAtLineBreak " << IndentAtLineBreak << "\n";
450     llvm::dbgs() << "DelimitersOnNewline " << DelimitersOnNewline << "\n";
451     for (size_t i = 0; i < Lines.size(); ++i) {
452       llvm::dbgs() << i << " |" << Content[i] << "| "
453                    << "CC=" << ContentColumn[i] << "| "
454                    << "IN=" << (Content[i].data() - Lines[i].data()) << "\n";
455     }
456   });
457 }
458 
459 BreakableToken::Split
460 BreakableBlockComment::getSplit(unsigned LineIndex, unsigned TailOffset,
461                            unsigned ColumnLimit, unsigned ContentStartColumn,
462                            llvm::Regex &CommentPragmasRegex) const {
463   // Don't break lines matching the comment pragmas regex.
464   if (CommentPragmasRegex.match(Content[LineIndex]))
465     return Split(StringRef::npos, 0);
466   return getCommentSplit(Content[LineIndex].substr(TailOffset),
467                          ContentStartColumn, ColumnLimit, Style.TabWidth,
468                          Encoding, Style, Decoration.endswith("*"));
469 }
470 
471 void BreakableBlockComment::adjustWhitespace(unsigned LineIndex,
472                                              int IndentDelta) {
473   // When in a preprocessor directive, the trailing backslash in a block comment
474   // is not needed, but can serve a purpose of uniformity with necessary escaped
475   // newlines outside the comment. In this case we remove it here before
476   // trimming the trailing whitespace. The backslash will be re-added later when
477   // inserting a line break.
478   size_t EndOfPreviousLine = Lines[LineIndex - 1].size();
479   if (InPPDirective && Lines[LineIndex - 1].endswith("\\"))
480     --EndOfPreviousLine;
481 
482   // Calculate the end of the non-whitespace text in the previous line.
483   EndOfPreviousLine =
484       Lines[LineIndex - 1].find_last_not_of(Blanks, EndOfPreviousLine);
485   if (EndOfPreviousLine == StringRef::npos)
486     EndOfPreviousLine = 0;
487   else
488     ++EndOfPreviousLine;
489   // Calculate the start of the non-whitespace text in the current line.
490   size_t StartOfLine = Lines[LineIndex].find_first_not_of(Blanks);
491   if (StartOfLine == StringRef::npos)
492     StartOfLine = Lines[LineIndex].size();
493 
494   StringRef Whitespace = Lines[LineIndex].substr(0, StartOfLine);
495   // Adjust Lines to only contain relevant text.
496   size_t PreviousContentOffset =
497       Content[LineIndex - 1].data() - Lines[LineIndex - 1].data();
498   Content[LineIndex - 1] = Lines[LineIndex - 1].substr(
499       PreviousContentOffset, EndOfPreviousLine - PreviousContentOffset);
500   Content[LineIndex] = Lines[LineIndex].substr(StartOfLine);
501 
502   // Adjust the start column uniformly across all lines.
503   ContentColumn[LineIndex] =
504       encoding::columnWidthWithTabs(Whitespace, 0, Style.TabWidth, Encoding) +
505       IndentDelta;
506 }
507 
508 unsigned BreakableBlockComment::getRangeLength(unsigned LineIndex,
509                                                unsigned Offset,
510                                                StringRef::size_type Length,
511                                                unsigned StartColumn) const {
512   unsigned LineLength =
513       encoding::columnWidthWithTabs(Content[LineIndex].substr(Offset, Length),
514                                     StartColumn, Style.TabWidth, Encoding);
515   // FIXME: This should go into getRemainingLength instead, but we currently
516   // break tests when putting it there. Investigate how to fix those tests.
517   // The last line gets a "*/" postfix.
518   if (LineIndex + 1 == Lines.size()) {
519     LineLength += 2;
520     // We never need a decoration when breaking just the trailing "*/" postfix.
521     // Note that checking that Length == 0 is not enough, since Length could
522     // also be StringRef::npos.
523     if (Content[LineIndex].substr(Offset, StringRef::npos).empty()) {
524       LineLength -= Decoration.size();
525     }
526   }
527   return LineLength;
528 }
529 
530 unsigned BreakableBlockComment::getRemainingLength(unsigned LineIndex,
531                                                    unsigned Offset,
532                                                    unsigned StartColumn) const {
533   return UnbreakableTailLength +
534          getRangeLength(LineIndex, Offset, StringRef::npos, StartColumn);
535 }
536 
537 unsigned BreakableBlockComment::getContentStartColumn(unsigned LineIndex,
538                                                       bool Break) const {
539   if (Break)
540     return IndentAtLineBreak;
541   return std::max(0, ContentColumn[LineIndex]);
542 }
543 
544 const llvm::StringSet<>
545     BreakableBlockComment::ContentIndentingJavadocAnnotations = {
546         "@param", "@return",     "@returns", "@throws",  "@type", "@template",
547         "@see",   "@deprecated", "@define",  "@exports", "@mods", "@private",
548 };
549 
550 unsigned BreakableBlockComment::getContentIndent(unsigned LineIndex) const {
551   if (Style.Language != FormatStyle::LK_Java &&
552       Style.Language != FormatStyle::LK_JavaScript)
553     return 0;
554   // The content at LineIndex 0 of a comment like:
555   // /** line 0 */
556   // is "* line 0", so we need to skip over the decoration in that case.
557   StringRef ContentWithNoDecoration = Content[LineIndex];
558   if (LineIndex == 0 && ContentWithNoDecoration.startswith("*")) {
559     ContentWithNoDecoration = ContentWithNoDecoration.substr(1).ltrim(Blanks);
560   }
561   StringRef FirstWord = ContentWithNoDecoration.substr(
562       0, ContentWithNoDecoration.find_first_of(Blanks));
563   if (ContentIndentingJavadocAnnotations.find(FirstWord) !=
564       ContentIndentingJavadocAnnotations.end())
565     return Style.ContinuationIndentWidth;
566   return 0;
567 }
568 
569 void BreakableBlockComment::insertBreak(unsigned LineIndex, unsigned TailOffset,
570                                         Split Split, unsigned ContentIndent,
571                                         WhitespaceManager &Whitespaces) const {
572   StringRef Text = Content[LineIndex].substr(TailOffset);
573   StringRef Prefix = Decoration;
574   // We need this to account for the case when we have a decoration "* " for all
575   // the lines except for the last one, where the star in "*/" acts as a
576   // decoration.
577   unsigned LocalIndentAtLineBreak = IndentAtLineBreak;
578   if (LineIndex + 1 == Lines.size() &&
579       Text.size() == Split.first + Split.second) {
580     // For the last line we need to break before "*/", but not to add "* ".
581     Prefix = "";
582     if (LocalIndentAtLineBreak >= 2)
583       LocalIndentAtLineBreak -= 2;
584   }
585   // The split offset is from the beginning of the line. Convert it to an offset
586   // from the beginning of the token text.
587   unsigned BreakOffsetInToken =
588       Text.data() - tokenAt(LineIndex).TokenText.data() + Split.first;
589   unsigned CharsToRemove = Split.second;
590   assert(LocalIndentAtLineBreak >= Prefix.size());
591   std::string PrefixWithTrailingIndent = Prefix;
592   for (unsigned I = 0; I < ContentIndent; ++I)
593     PrefixWithTrailingIndent += " ";
594   Whitespaces.replaceWhitespaceInToken(
595       tokenAt(LineIndex), BreakOffsetInToken, CharsToRemove, "",
596       PrefixWithTrailingIndent, InPPDirective, /*Newlines=*/1,
597       /*Spaces=*/LocalIndentAtLineBreak + ContentIndent -
598           PrefixWithTrailingIndent.size());
599 }
600 
601 BreakableToken::Split
602 BreakableBlockComment::getReflowSplit(unsigned LineIndex,
603                                       llvm::Regex &CommentPragmasRegex) const {
604   if (!mayReflow(LineIndex, CommentPragmasRegex))
605     return Split(StringRef::npos, 0);
606 
607   // If we're reflowing into a line with content indent, only reflow the next
608   // line if its starting whitespace matches the content indent.
609   size_t Trimmed = Content[LineIndex].find_first_not_of(Blanks);
610   if (LineIndex) {
611     unsigned PreviousContentIndent = getContentIndent(LineIndex - 1);
612     if (PreviousContentIndent && Trimmed != StringRef::npos &&
613         Trimmed != PreviousContentIndent)
614       return Split(StringRef::npos, 0);
615   }
616 
617   return Split(0, Trimmed != StringRef::npos ? Trimmed : 0);
618 }
619 
620 bool BreakableBlockComment::introducesBreakBeforeToken() const {
621   // A break is introduced when we want delimiters on newline.
622   return DelimitersOnNewline &&
623          Lines[0].substr(1).find_first_not_of(Blanks) != StringRef::npos;
624 }
625 
626 void BreakableBlockComment::reflow(unsigned LineIndex,
627                                    WhitespaceManager &Whitespaces) const {
628   StringRef TrimmedContent = Content[LineIndex].ltrim(Blanks);
629   // Here we need to reflow.
630   assert(Tokens[LineIndex - 1] == Tokens[LineIndex] &&
631          "Reflowing whitespace within a token");
632   // This is the offset of the end of the last line relative to the start of
633   // the token text in the token.
634   unsigned WhitespaceOffsetInToken = Content[LineIndex - 1].data() +
635                                      Content[LineIndex - 1].size() -
636                                      tokenAt(LineIndex).TokenText.data();
637   unsigned WhitespaceLength = TrimmedContent.data() -
638                               tokenAt(LineIndex).TokenText.data() -
639                               WhitespaceOffsetInToken;
640   Whitespaces.replaceWhitespaceInToken(
641       tokenAt(LineIndex), WhitespaceOffsetInToken,
642       /*ReplaceChars=*/WhitespaceLength, /*PreviousPostfix=*/"",
643       /*CurrentPrefix=*/ReflowPrefix, InPPDirective, /*Newlines=*/0,
644       /*Spaces=*/0);
645 }
646 
647 void BreakableBlockComment::adaptStartOfLine(
648     unsigned LineIndex, WhitespaceManager &Whitespaces) const {
649   if (LineIndex == 0) {
650     if (DelimitersOnNewline) {
651       // Since we're breaking at index 1 below, the break position and the
652       // break length are the same.
653       // Note: this works because getCommentSplit is careful never to split at
654       // the beginning of a line.
655       size_t BreakLength = Lines[0].substr(1).find_first_not_of(Blanks);
656       if (BreakLength != StringRef::npos)
657         insertBreak(LineIndex, 0, Split(1, BreakLength), /*ContentIndent=*/0,
658                     Whitespaces);
659     }
660     return;
661   }
662   // Here no reflow with the previous line will happen.
663   // Fix the decoration of the line at LineIndex.
664   StringRef Prefix = Decoration;
665   if (Content[LineIndex].empty()) {
666     if (LineIndex + 1 == Lines.size()) {
667       if (!LastLineNeedsDecoration) {
668         // If the last line was empty, we don't need a prefix, as the */ will
669         // line up with the decoration (if it exists).
670         Prefix = "";
671       }
672     } else if (!Decoration.empty()) {
673       // For other empty lines, if we do have a decoration, adapt it to not
674       // contain a trailing whitespace.
675       Prefix = Prefix.substr(0, 1);
676     }
677   } else {
678     if (ContentColumn[LineIndex] == 1) {
679       // This line starts immediately after the decorating *.
680       Prefix = Prefix.substr(0, 1);
681     }
682   }
683   // This is the offset of the end of the last line relative to the start of the
684   // token text in the token.
685   unsigned WhitespaceOffsetInToken = Content[LineIndex - 1].data() +
686                                      Content[LineIndex - 1].size() -
687                                      tokenAt(LineIndex).TokenText.data();
688   unsigned WhitespaceLength = Content[LineIndex].data() -
689                               tokenAt(LineIndex).TokenText.data() -
690                               WhitespaceOffsetInToken;
691   Whitespaces.replaceWhitespaceInToken(
692       tokenAt(LineIndex), WhitespaceOffsetInToken, WhitespaceLength, "", Prefix,
693       InPPDirective, /*Newlines=*/1, ContentColumn[LineIndex] - Prefix.size());
694 }
695 
696 BreakableToken::Split
697 BreakableBlockComment::getSplitAfterLastLine(unsigned TailOffset) const {
698   if (DelimitersOnNewline) {
699     // Replace the trailing whitespace of the last line with a newline.
700     // In case the last line is empty, the ending '*/' is already on its own
701     // line.
702     StringRef Line = Content.back().substr(TailOffset);
703     StringRef TrimmedLine = Line.rtrim(Blanks);
704     if (!TrimmedLine.empty())
705       return Split(TrimmedLine.size(), Line.size() - TrimmedLine.size());
706   }
707   return Split(StringRef::npos, 0);
708 }
709 
710 bool BreakableBlockComment::mayReflow(unsigned LineIndex,
711                                       llvm::Regex &CommentPragmasRegex) const {
712   // Content[LineIndex] may exclude the indent after the '*' decoration. In that
713   // case, we compute the start of the comment pragma manually.
714   StringRef IndentContent = Content[LineIndex];
715   if (Lines[LineIndex].ltrim(Blanks).startswith("*")) {
716     IndentContent = Lines[LineIndex].ltrim(Blanks).substr(1);
717   }
718   return LineIndex > 0 && !CommentPragmasRegex.match(IndentContent) &&
719          mayReflowContent(Content[LineIndex]) && !Tok.Finalized &&
720          !switchesFormatting(tokenAt(LineIndex));
721 }
722 
723 BreakableLineCommentSection::BreakableLineCommentSection(
724     const FormatToken &Token, unsigned StartColumn,
725     unsigned OriginalStartColumn, bool FirstInLine, bool InPPDirective,
726     encoding::Encoding Encoding, const FormatStyle &Style)
727     : BreakableComment(Token, StartColumn, InPPDirective, Encoding, Style) {
728   assert(Tok.is(TT_LineComment) &&
729          "line comment section must start with a line comment");
730   FormatToken *LineTok = nullptr;
731   for (const FormatToken *CurrentTok = &Tok;
732        CurrentTok && CurrentTok->is(TT_LineComment);
733        CurrentTok = CurrentTok->Next) {
734     LastLineTok = LineTok;
735     StringRef TokenText(CurrentTok->TokenText);
736     assert((TokenText.startswith("//") || TokenText.startswith("#")) &&
737            "unsupported line comment prefix, '//' and '#' are supported");
738     size_t FirstLineIndex = Lines.size();
739     TokenText.split(Lines, "\n");
740     Content.resize(Lines.size());
741     ContentColumn.resize(Lines.size());
742     OriginalContentColumn.resize(Lines.size());
743     Tokens.resize(Lines.size());
744     Prefix.resize(Lines.size());
745     OriginalPrefix.resize(Lines.size());
746     for (size_t i = FirstLineIndex, e = Lines.size(); i < e; ++i) {
747       Lines[i] = Lines[i].ltrim(Blanks);
748       // We need to trim the blanks in case this is not the first line in a
749       // multiline comment. Then the indent is included in Lines[i].
750       StringRef IndentPrefix =
751           getLineCommentIndentPrefix(Lines[i].ltrim(Blanks), Style);
752       assert((TokenText.startswith("//") || TokenText.startswith("#")) &&
753              "unsupported line comment prefix, '//' and '#' are supported");
754       OriginalPrefix[i] = Prefix[i] = IndentPrefix;
755       if (Lines[i].size() > Prefix[i].size() &&
756           isAlphanumeric(Lines[i][Prefix[i].size()])) {
757         if (Prefix[i] == "//")
758           Prefix[i] = "// ";
759         else if (Prefix[i] == "///")
760           Prefix[i] = "/// ";
761         else if (Prefix[i] == "//!")
762           Prefix[i] = "//! ";
763         else if (Prefix[i] == "///<")
764           Prefix[i] = "///< ";
765         else if (Prefix[i] == "//!<")
766           Prefix[i] = "//!< ";
767         else if (Prefix[i] == "#" &&
768                  Style.Language == FormatStyle::LK_TextProto)
769           Prefix[i] = "# ";
770       }
771 
772       Tokens[i] = LineTok;
773       Content[i] = Lines[i].substr(IndentPrefix.size());
774       OriginalContentColumn[i] =
775           StartColumn + encoding::columnWidthWithTabs(OriginalPrefix[i],
776                                                       StartColumn,
777                                                       Style.TabWidth, Encoding);
778       ContentColumn[i] =
779           StartColumn + encoding::columnWidthWithTabs(Prefix[i], StartColumn,
780                                                       Style.TabWidth, Encoding);
781 
782       // Calculate the end of the non-whitespace text in this line.
783       size_t EndOfLine = Content[i].find_last_not_of(Blanks);
784       if (EndOfLine == StringRef::npos)
785         EndOfLine = Content[i].size();
786       else
787         ++EndOfLine;
788       Content[i] = Content[i].substr(0, EndOfLine);
789     }
790     LineTok = CurrentTok->Next;
791     if (CurrentTok->Next && !CurrentTok->Next->ContinuesLineCommentSection) {
792       // A line comment section needs to broken by a line comment that is
793       // preceded by at least two newlines. Note that we put this break here
794       // instead of breaking at a previous stage during parsing, since that
795       // would split the contents of the enum into two unwrapped lines in this
796       // example, which is undesirable:
797       // enum A {
798       //   a, // comment about a
799       //
800       //   // comment about b
801       //   b
802       // };
803       //
804       // FIXME: Consider putting separate line comment sections as children to
805       // the unwrapped line instead.
806       break;
807     }
808   }
809 }
810 
811 unsigned
812 BreakableLineCommentSection::getRangeLength(unsigned LineIndex, unsigned Offset,
813                                             StringRef::size_type Length,
814                                             unsigned StartColumn) const {
815   return encoding::columnWidthWithTabs(
816       Content[LineIndex].substr(Offset, Length), StartColumn, Style.TabWidth,
817       Encoding);
818 }
819 
820 unsigned BreakableLineCommentSection::getContentStartColumn(unsigned LineIndex,
821                                                             bool Break) const {
822   if (Break)
823     return OriginalContentColumn[LineIndex];
824   return ContentColumn[LineIndex];
825 }
826 
827 void BreakableLineCommentSection::insertBreak(
828     unsigned LineIndex, unsigned TailOffset, Split Split,
829     unsigned ContentIndent, WhitespaceManager &Whitespaces) const {
830   StringRef Text = Content[LineIndex].substr(TailOffset);
831   // Compute the offset of the split relative to the beginning of the token
832   // text.
833   unsigned BreakOffsetInToken =
834       Text.data() - tokenAt(LineIndex).TokenText.data() + Split.first;
835   unsigned CharsToRemove = Split.second;
836   // Compute the size of the new indent, including the size of the new prefix of
837   // the newly broken line.
838   unsigned IndentAtLineBreak = OriginalContentColumn[LineIndex] +
839                                Prefix[LineIndex].size() -
840                                OriginalPrefix[LineIndex].size();
841   assert(IndentAtLineBreak >= Prefix[LineIndex].size());
842   Whitespaces.replaceWhitespaceInToken(
843       tokenAt(LineIndex), BreakOffsetInToken, CharsToRemove, "",
844       Prefix[LineIndex], InPPDirective, /*Newlines=*/1,
845       /*Spaces=*/IndentAtLineBreak - Prefix[LineIndex].size());
846 }
847 
848 BreakableComment::Split BreakableLineCommentSection::getReflowSplit(
849     unsigned LineIndex, llvm::Regex &CommentPragmasRegex) const {
850   if (!mayReflow(LineIndex, CommentPragmasRegex))
851     return Split(StringRef::npos, 0);
852 
853   size_t Trimmed = Content[LineIndex].find_first_not_of(Blanks);
854 
855   // In a line comment section each line is a separate token; thus, after a
856   // split we replace all whitespace before the current line comment token
857   // (which does not need to be included in the split), plus the start of the
858   // line up to where the content starts.
859   return Split(0, Trimmed != StringRef::npos ? Trimmed : 0);
860 }
861 
862 void BreakableLineCommentSection::reflow(unsigned LineIndex,
863                                          WhitespaceManager &Whitespaces) const {
864   if (LineIndex > 0 && Tokens[LineIndex] != Tokens[LineIndex - 1]) {
865     // Reflow happens between tokens. Replace the whitespace between the
866     // tokens by the empty string.
867     Whitespaces.replaceWhitespace(
868         *Tokens[LineIndex], /*Newlines=*/0, /*Spaces=*/0,
869         /*StartOfTokenColumn=*/StartColumn, /*InPPDirective=*/false);
870   } else if (LineIndex > 0) {
871     // In case we're reflowing after the '\' in:
872     //
873     //   // line comment \
874     //   // line 2
875     //
876     // the reflow happens inside the single comment token (it is a single line
877     // comment with an unescaped newline).
878     // Replace the whitespace between the '\' and '//' with the empty string.
879     //
880     // Offset points to after the '\' relative to start of the token.
881     unsigned Offset = Lines[LineIndex - 1].data() +
882                       Lines[LineIndex - 1].size() -
883                       tokenAt(LineIndex - 1).TokenText.data();
884     // WhitespaceLength is the number of chars between the '\' and the '//' on
885     // the next line.
886     unsigned WhitespaceLength =
887         Lines[LineIndex].data() - tokenAt(LineIndex).TokenText.data() - Offset;
888     Whitespaces.replaceWhitespaceInToken(*Tokens[LineIndex], Offset,
889                                          /*ReplaceChars=*/WhitespaceLength,
890                                          /*PreviousPostfix=*/"",
891                                          /*CurrentPrefix=*/"",
892                                          /*InPPDirective=*/false,
893                                          /*Newlines=*/0,
894                                          /*Spaces=*/0);
895   }
896   // Replace the indent and prefix of the token with the reflow prefix.
897   unsigned Offset =
898       Lines[LineIndex].data() - tokenAt(LineIndex).TokenText.data();
899   unsigned WhitespaceLength =
900       Content[LineIndex].data() - Lines[LineIndex].data();
901   Whitespaces.replaceWhitespaceInToken(*Tokens[LineIndex], Offset,
902                                        /*ReplaceChars=*/WhitespaceLength,
903                                        /*PreviousPostfix=*/"",
904                                        /*CurrentPrefix=*/ReflowPrefix,
905                                        /*InPPDirective=*/false,
906                                        /*Newlines=*/0,
907                                        /*Spaces=*/0);
908 }
909 
910 void BreakableLineCommentSection::adaptStartOfLine(
911     unsigned LineIndex, WhitespaceManager &Whitespaces) const {
912   // If this is the first line of a token, we need to inform Whitespace Manager
913   // about it: either adapt the whitespace range preceding it, or mark it as an
914   // untouchable token.
915   // This happens for instance here:
916   // // line 1 \
917   // // line 2
918   if (LineIndex > 0 && Tokens[LineIndex] != Tokens[LineIndex - 1]) {
919     // This is the first line for the current token, but no reflow with the
920     // previous token is necessary. However, we still may need to adjust the
921     // start column. Note that ContentColumn[LineIndex] is the expected
922     // content column after a possible update to the prefix, hence the prefix
923     // length change is included.
924     unsigned LineColumn =
925         ContentColumn[LineIndex] -
926         (Content[LineIndex].data() - Lines[LineIndex].data()) +
927         (OriginalPrefix[LineIndex].size() - Prefix[LineIndex].size());
928 
929     // We always want to create a replacement instead of adding an untouchable
930     // token, even if LineColumn is the same as the original column of the
931     // token. This is because WhitespaceManager doesn't align trailing
932     // comments if they are untouchable.
933     Whitespaces.replaceWhitespace(*Tokens[LineIndex],
934                                   /*Newlines=*/1,
935                                   /*Spaces=*/LineColumn,
936                                   /*StartOfTokenColumn=*/LineColumn,
937                                   /*InPPDirective=*/false);
938   }
939   if (OriginalPrefix[LineIndex] != Prefix[LineIndex]) {
940     // Adjust the prefix if necessary.
941 
942     // Take care of the space possibly introduced after a decoration.
943     assert(Prefix[LineIndex] == (OriginalPrefix[LineIndex] + " ").str() &&
944            "Expecting a line comment prefix to differ from original by at most "
945            "a space");
946     Whitespaces.replaceWhitespaceInToken(
947         tokenAt(LineIndex), OriginalPrefix[LineIndex].size(), 0, "", "",
948         /*InPPDirective=*/false, /*Newlines=*/0, /*Spaces=*/1);
949   }
950 }
951 
952 void BreakableLineCommentSection::updateNextToken(LineState &State) const {
953   if (LastLineTok) {
954     State.NextToken = LastLineTok->Next;
955   }
956 }
957 
958 bool BreakableLineCommentSection::mayReflow(
959     unsigned LineIndex, llvm::Regex &CommentPragmasRegex) const {
960   // Line comments have the indent as part of the prefix, so we need to
961   // recompute the start of the line.
962   StringRef IndentContent = Content[LineIndex];
963   if (Lines[LineIndex].startswith("//")) {
964     IndentContent = Lines[LineIndex].substr(2);
965   }
966   // FIXME: Decide whether we want to reflow non-regular indents:
967   // Currently, we only reflow when the OriginalPrefix[LineIndex] matches the
968   // OriginalPrefix[LineIndex-1]. That means we don't reflow
969   // // text that protrudes
970   // //    into text with different indent
971   // We do reflow in that case in block comments.
972   return LineIndex > 0 && !CommentPragmasRegex.match(IndentContent) &&
973          mayReflowContent(Content[LineIndex]) && !Tok.Finalized &&
974          !switchesFormatting(tokenAt(LineIndex)) &&
975          OriginalPrefix[LineIndex] == OriginalPrefix[LineIndex - 1];
976 }
977 
978 } // namespace format
979 } // namespace clang
980