1 //===--- BreakableToken.cpp - Format C++ code -----------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
11 /// \brief Contains implementation of BreakableToken class and classes derived
14 //===----------------------------------------------------------------------===//
16 #include "BreakableToken.h"
17 #include "ContinuationIndenter.h"
18 #include "clang/Basic/CharInfo.h"
19 #include "clang/Format/Format.h"
20 #include "llvm/ADT/STLExtras.h"
21 #include "llvm/Support/Debug.h"
24 #define DEBUG_TYPE "format-token-breaker"
29 static const char *const Blanks = " \t\v\f\r";
30 static bool IsBlank(char C) {
43 static StringRef getLineCommentIndentPrefix(StringRef Comment,
44 const FormatStyle &Style) {
45 static const char *const KnownCStylePrefixes[] = {"///<", "//!<", "///", "//",
47 static const char *const KnownTextProtoPrefixes[] = {"//", "#"};
48 ArrayRef<const char *> KnownPrefixes(KnownCStylePrefixes);
49 if (Style.Language == FormatStyle::LK_TextProto)
50 KnownPrefixes = KnownTextProtoPrefixes;
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] == ' ')
58 if (PrefixLength > LongestPrefix.size())
59 LongestPrefix = Comment.substr(0, PrefixLength);
65 static BreakableToken::Split getCommentSplit(StringRef Text,
66 unsigned ContentStartColumn,
69 encoding::Encoding Encoding) {
70 DEBUG(llvm::dbgs() << "Comment split: \"" << Text << ", " << ColumnLimit
71 << "\", Content start: " << ContentStartColumn << "\n");
72 if (ColumnLimit <= ContentStartColumn + 1)
73 return BreakableToken::Split(StringRef::npos, 0);
75 unsigned MaxSplit = ColumnLimit - ContentStartColumn + 1;
76 unsigned MaxSplitBytes = 0;
78 for (unsigned NumChars = 0;
79 NumChars < MaxSplit && MaxSplitBytes < Text.size();) {
80 unsigned BytesInChar =
81 encoding::getCodePointNumBytes(Text[MaxSplitBytes], Encoding);
83 encoding::columnWidthWithTabs(Text.substr(MaxSplitBytes, BytesInChar),
84 ContentStartColumn, TabWidth, Encoding);
85 MaxSplitBytes += BytesInChar;
88 StringRef::size_type SpaceOffset = Text.find_last_of(Blanks, MaxSplitBytes);
90 // Do not split before a number followed by a dot: this would be interpreted
91 // as a numbered list, which would prevent re-flowing in subsequent passes.
92 static llvm::Regex kNumberedListRegexp = llvm::Regex("^[1-9][0-9]?\\.");
93 if (SpaceOffset != StringRef::npos &&
94 kNumberedListRegexp.match(Text.substr(SpaceOffset).ltrim(Blanks)))
95 SpaceOffset = Text.find_last_of(Blanks, SpaceOffset);
97 if (SpaceOffset == StringRef::npos ||
98 // Don't break at leading whitespace.
99 Text.find_last_not_of(Blanks, SpaceOffset) == StringRef::npos) {
100 // Make sure that we don't break at leading whitespace that
101 // reaches past MaxSplit.
102 StringRef::size_type FirstNonWhitespace = Text.find_first_not_of(Blanks);
103 if (FirstNonWhitespace == StringRef::npos)
104 // If the comment is only whitespace, we cannot split.
105 return BreakableToken::Split(StringRef::npos, 0);
106 SpaceOffset = Text.find_first_of(
107 Blanks, std::max<unsigned>(MaxSplitBytes, FirstNonWhitespace));
109 if (SpaceOffset != StringRef::npos && SpaceOffset != 0) {
110 StringRef BeforeCut = Text.substr(0, SpaceOffset).rtrim(Blanks);
111 StringRef AfterCut = Text.substr(SpaceOffset).ltrim(Blanks);
112 return BreakableToken::Split(BeforeCut.size(),
113 AfterCut.begin() - BeforeCut.end());
115 return BreakableToken::Split(StringRef::npos, 0);
118 static BreakableToken::Split
119 getStringSplit(StringRef Text, unsigned UsedColumns, unsigned ColumnLimit,
120 unsigned TabWidth, encoding::Encoding Encoding) {
121 // FIXME: Reduce unit test case.
123 return BreakableToken::Split(StringRef::npos, 0);
124 if (ColumnLimit <= UsedColumns)
125 return BreakableToken::Split(StringRef::npos, 0);
126 unsigned MaxSplit = ColumnLimit - UsedColumns;
127 StringRef::size_type SpaceOffset = 0;
128 StringRef::size_type SlashOffset = 0;
129 StringRef::size_type WordStartOffset = 0;
130 StringRef::size_type SplitPoint = 0;
131 for (unsigned Chars = 0;;) {
133 if (Text[0] == '\\') {
134 Advance = encoding::getEscapeSequenceLength(Text);
137 Advance = encoding::getCodePointNumBytes(Text[0], Encoding);
138 Chars += encoding::columnWidthWithTabs(
139 Text.substr(0, Advance), UsedColumns + Chars, TabWidth, Encoding);
142 if (Chars > MaxSplit || Text.size() <= Advance)
145 if (IsBlank(Text[0]))
146 SpaceOffset = SplitPoint;
148 SlashOffset = SplitPoint;
149 if (Advance == 1 && !isAlphanumeric(Text[0]))
150 WordStartOffset = SplitPoint;
152 SplitPoint += Advance;
153 Text = Text.substr(Advance);
156 if (SpaceOffset != 0)
157 return BreakableToken::Split(SpaceOffset + 1, 0);
158 if (SlashOffset != 0)
159 return BreakableToken::Split(SlashOffset + 1, 0);
160 if (WordStartOffset != 0)
161 return BreakableToken::Split(WordStartOffset + 1, 0);
163 return BreakableToken::Split(SplitPoint, 0);
164 return BreakableToken::Split(StringRef::npos, 0);
167 bool switchesFormatting(const FormatToken &Token) {
168 assert((Token.is(TT_BlockComment) || Token.is(TT_LineComment)) &&
169 "formatting regions are switched by comment tokens");
170 StringRef Content = Token.TokenText.substr(2).ltrim();
171 return Content.startswith("clang-format on") ||
172 Content.startswith("clang-format off");
176 BreakableToken::getLengthAfterCompression(unsigned RemainingTokenColumns,
178 // Example: consider the content
180 // - RemainingTokenColumns is the original number of columns, 10;
181 // - Split is (4, 2), denoting the two spaces between the two words;
183 // We compute the number of columns when the split is compressed into a single
187 // FIXME: Correctly measure the length of whitespace in Split.second so it
189 return RemainingTokenColumns + 1 - Split.second;
192 unsigned BreakableStringLiteral::getLineCount() const { return 1; }
194 unsigned BreakableStringLiteral::getRangeLength(unsigned LineIndex,
196 StringRef::size_type Length,
197 unsigned StartColumn) const {
198 llvm_unreachable("Getting the length of a part of the string literal "
199 "indicates that the code tries to reflow it.");
203 BreakableStringLiteral::getRemainingLength(unsigned LineIndex, unsigned Offset,
204 unsigned StartColumn) const {
205 return UnbreakableTailLength + Postfix.size() +
206 encoding::columnWidthWithTabs(Line.substr(Offset, StringRef::npos),
207 StartColumn, Style.TabWidth, Encoding);
210 unsigned BreakableStringLiteral::getContentStartColumn(unsigned LineIndex,
212 return StartColumn + Prefix.size();
215 BreakableStringLiteral::BreakableStringLiteral(
216 const FormatToken &Tok, unsigned StartColumn, StringRef Prefix,
217 StringRef Postfix, bool InPPDirective, encoding::Encoding Encoding,
218 const FormatStyle &Style)
219 : BreakableToken(Tok, InPPDirective, Encoding, Style),
220 StartColumn(StartColumn), Prefix(Prefix), Postfix(Postfix),
221 UnbreakableTailLength(Tok.UnbreakableTailLength) {
222 assert(Tok.TokenText.startswith(Prefix) && Tok.TokenText.endswith(Postfix));
223 Line = Tok.TokenText.substr(
224 Prefix.size(), Tok.TokenText.size() - Prefix.size() - Postfix.size());
227 BreakableToken::Split BreakableStringLiteral::getSplit(
228 unsigned LineIndex, unsigned TailOffset, unsigned ColumnLimit,
229 unsigned ContentStartColumn, llvm::Regex &CommentPragmasRegex) const {
230 return getStringSplit(Line.substr(TailOffset), ContentStartColumn,
231 ColumnLimit - Postfix.size(), Style.TabWidth, Encoding);
234 void BreakableStringLiteral::insertBreak(unsigned LineIndex,
235 unsigned TailOffset, Split Split,
236 WhitespaceManager &Whitespaces) const {
237 Whitespaces.replaceWhitespaceInToken(
238 Tok, Prefix.size() + TailOffset + Split.first, Split.second, Postfix,
239 Prefix, InPPDirective, 1, StartColumn);
242 BreakableComment::BreakableComment(const FormatToken &Token,
243 unsigned StartColumn, bool InPPDirective,
244 encoding::Encoding Encoding,
245 const FormatStyle &Style)
246 : BreakableToken(Token, InPPDirective, Encoding, Style),
247 StartColumn(StartColumn) {}
249 unsigned BreakableComment::getLineCount() const { return Lines.size(); }
251 BreakableToken::Split
252 BreakableComment::getSplit(unsigned LineIndex, unsigned TailOffset,
253 unsigned ColumnLimit, unsigned ContentStartColumn,
254 llvm::Regex &CommentPragmasRegex) const {
255 // Don't break lines matching the comment pragmas regex.
256 if (CommentPragmasRegex.match(Content[LineIndex]))
257 return Split(StringRef::npos, 0);
258 return getCommentSplit(Content[LineIndex].substr(TailOffset),
259 ContentStartColumn, ColumnLimit, Style.TabWidth,
263 void BreakableComment::compressWhitespace(
264 unsigned LineIndex, unsigned TailOffset, Split Split,
265 WhitespaceManager &Whitespaces) const {
266 StringRef Text = Content[LineIndex].substr(TailOffset);
267 // Text is relative to the content line, but Whitespaces operates relative to
268 // the start of the corresponding token, so compute the start of the Split
269 // that needs to be compressed into a single space relative to the start of
271 unsigned BreakOffsetInToken =
272 Text.data() - tokenAt(LineIndex).TokenText.data() + Split.first;
273 unsigned CharsToRemove = Split.second;
274 Whitespaces.replaceWhitespaceInToken(
275 tokenAt(LineIndex), BreakOffsetInToken, CharsToRemove, "", "",
276 /*InPPDirective=*/false, /*Newlines=*/0, /*Spaces=*/1);
279 const FormatToken &BreakableComment::tokenAt(unsigned LineIndex) const {
280 return Tokens[LineIndex] ? *Tokens[LineIndex] : Tok;
283 static bool mayReflowContent(StringRef Content) {
284 Content = Content.trim(Blanks);
285 // Lines starting with '@' commonly have special meaning.
286 // Lines starting with '-', '-#', '+' or '*' are bulleted/numbered lists.
287 static const SmallVector<StringRef, 8> kSpecialMeaningPrefixes = {
288 "@", "TODO", "FIXME", "XXX", "-# ", "- ", "+ ", "* "};
289 bool hasSpecialMeaningPrefix = false;
290 for (StringRef Prefix : kSpecialMeaningPrefixes) {
291 if (Content.startswith(Prefix)) {
292 hasSpecialMeaningPrefix = true;
297 // Numbered lists may also start with a number followed by '.'
298 // To avoid issues if a line starts with a number which is actually the end
299 // of a previous line, we only consider numbers with up to 2 digits.
300 static llvm::Regex kNumberedListRegexp = llvm::Regex("^[1-9][0-9]?\\. ");
301 hasSpecialMeaningPrefix =
302 hasSpecialMeaningPrefix || kNumberedListRegexp.match(Content);
304 // Simple heuristic for what to reflow: content should contain at least two
305 // characters and either the first or second character must be
307 return Content.size() >= 2 && !hasSpecialMeaningPrefix &&
308 !Content.endswith("\\") &&
309 // Note that this is UTF-8 safe, since if isPunctuation(Content[0]) is
310 // true, then the first code point must be 1 byte long.
311 (!isPunctuation(Content[0]) || !isPunctuation(Content[1]));
314 BreakableBlockComment::BreakableBlockComment(
315 const FormatToken &Token, unsigned StartColumn,
316 unsigned OriginalStartColumn, bool FirstInLine, bool InPPDirective,
317 encoding::Encoding Encoding, const FormatStyle &Style)
318 : BreakableComment(Token, StartColumn, InPPDirective, Encoding, Style),
319 DelimitersOnNewline(false),
320 UnbreakableTailLength(Token.UnbreakableTailLength) {
321 assert(Tok.is(TT_BlockComment) &&
322 "block comment section must start with a block comment");
324 StringRef TokenText(Tok.TokenText);
325 assert(TokenText.startswith("/*") && TokenText.endswith("*/"));
326 TokenText.substr(2, TokenText.size() - 4).split(Lines, "\n");
328 int IndentDelta = StartColumn - OriginalStartColumn;
329 Content.resize(Lines.size());
330 Content[0] = Lines[0];
331 ContentColumn.resize(Lines.size());
332 // Account for the initial '/*'.
333 ContentColumn[0] = StartColumn + 2;
334 Tokens.resize(Lines.size());
335 for (size_t i = 1; i < Lines.size(); ++i)
336 adjustWhitespace(i, IndentDelta);
338 // Align decorations with the column of the star on the first line,
339 // that is one column after the start "/*".
340 DecorationColumn = StartColumn + 1;
342 // Account for comment decoration patterns like this:
347 if (Lines.size() >= 2 && Content[1].startswith("**") &&
348 static_cast<unsigned>(ContentColumn[1]) == StartColumn) {
349 DecorationColumn = StartColumn;
353 if (Lines.size() == 1 && !FirstInLine) {
354 // Comments for which FirstInLine is false can start on arbitrary column,
355 // and available horizontal space can be too small to align consecutive
356 // lines with the first one.
357 // FIXME: We could, probably, align them to current indentation level, but
358 // now we just wrap them without stars.
361 for (size_t i = 1, e = Lines.size(); i < e && !Decoration.empty(); ++i) {
362 // If the last line is empty, the closing "*/" will have a star.
363 if (i + 1 == e && Content[i].empty())
365 if (!Content[i].empty() && i + 1 != e && Decoration.startswith(Content[i]))
367 while (!Content[i].startswith(Decoration))
368 Decoration = Decoration.substr(0, Decoration.size() - 1);
371 LastLineNeedsDecoration = true;
372 IndentAtLineBreak = ContentColumn[0] + 1;
373 for (size_t i = 1, e = Lines.size(); i < e; ++i) {
374 if (Content[i].empty()) {
376 // Empty last line means that we already have a star as a part of the
377 // trailing */. We also need to preserve whitespace, so that */ is
378 // correctly indented.
379 LastLineNeedsDecoration = false;
380 // Align the star in the last '*/' with the stars on the previous lines.
381 if (e >= 2 && !Decoration.empty()) {
382 ContentColumn[i] = DecorationColumn;
384 } else if (Decoration.empty()) {
385 // For all other lines, set the start column to 0 if they're empty, so
386 // we do not insert trailing whitespace anywhere.
387 ContentColumn[i] = 0;
392 // The first line already excludes the star.
393 // The last line excludes the star if LastLineNeedsDecoration is false.
394 // For all other lines, adjust the line to exclude the star and
395 // (optionally) the first whitespace.
396 unsigned DecorationSize = Decoration.startswith(Content[i])
399 if (DecorationSize) {
400 ContentColumn[i] = DecorationColumn + DecorationSize;
402 Content[i] = Content[i].substr(DecorationSize);
403 if (!Decoration.startswith(Content[i]))
405 std::min<int>(IndentAtLineBreak, std::max(0, ContentColumn[i]));
407 IndentAtLineBreak = std::max<unsigned>(IndentAtLineBreak, Decoration.size());
409 // Detect a multiline jsdoc comment and set DelimitersOnNewline in that case.
410 if (Style.Language == FormatStyle::LK_JavaScript ||
411 Style.Language == FormatStyle::LK_Java) {
412 if ((Lines[0] == "*" || Lines[0].startswith("* ")) && Lines.size() > 1) {
413 // This is a multiline jsdoc comment.
414 DelimitersOnNewline = true;
415 } else if (Lines[0].startswith("* ") && Lines.size() == 1) {
416 // Detect a long single-line comment, like:
417 // /** long long long */
418 // Below, '2' is the width of '*/'.
421 encoding::columnWidthWithTabs(Lines[0], ContentColumn[0],
422 Style.TabWidth, Encoding) +
424 DelimitersOnNewline = EndColumn > Style.ColumnLimit;
429 llvm::dbgs() << "IndentAtLineBreak " << IndentAtLineBreak << "\n";
430 llvm::dbgs() << "DelimitersOnNewline " << DelimitersOnNewline << "\n";
431 for (size_t i = 0; i < Lines.size(); ++i) {
432 llvm::dbgs() << i << " |" << Content[i] << "| "
433 << "CC=" << ContentColumn[i] << "| "
434 << "IN=" << (Content[i].data() - Lines[i].data()) << "\n";
439 void BreakableBlockComment::adjustWhitespace(unsigned LineIndex,
441 // When in a preprocessor directive, the trailing backslash in a block comment
442 // is not needed, but can serve a purpose of uniformity with necessary escaped
443 // newlines outside the comment. In this case we remove it here before
444 // trimming the trailing whitespace. The backslash will be re-added later when
445 // inserting a line break.
446 size_t EndOfPreviousLine = Lines[LineIndex - 1].size();
447 if (InPPDirective && Lines[LineIndex - 1].endswith("\\"))
450 // Calculate the end of the non-whitespace text in the previous line.
452 Lines[LineIndex - 1].find_last_not_of(Blanks, EndOfPreviousLine);
453 if (EndOfPreviousLine == StringRef::npos)
454 EndOfPreviousLine = 0;
457 // Calculate the start of the non-whitespace text in the current line.
458 size_t StartOfLine = Lines[LineIndex].find_first_not_of(Blanks);
459 if (StartOfLine == StringRef::npos)
460 StartOfLine = Lines[LineIndex].rtrim("\r\n").size();
462 StringRef Whitespace = Lines[LineIndex].substr(0, StartOfLine);
463 // Adjust Lines to only contain relevant text.
464 size_t PreviousContentOffset =
465 Content[LineIndex - 1].data() - Lines[LineIndex - 1].data();
466 Content[LineIndex - 1] = Lines[LineIndex - 1].substr(
467 PreviousContentOffset, EndOfPreviousLine - PreviousContentOffset);
468 Content[LineIndex] = Lines[LineIndex].substr(StartOfLine);
470 // Adjust the start column uniformly across all lines.
471 ContentColumn[LineIndex] =
472 encoding::columnWidthWithTabs(Whitespace, 0, Style.TabWidth, Encoding) +
476 unsigned BreakableBlockComment::getRangeLength(unsigned LineIndex,
478 StringRef::size_type Length,
479 unsigned StartColumn) const {
480 unsigned LineLength =
481 encoding::columnWidthWithTabs(Content[LineIndex].substr(Offset, Length),
482 StartColumn, Style.TabWidth, Encoding);
483 // FIXME: This should go into getRemainingLength instead, but we currently
484 // break tests when putting it there. Investigate how to fix those tests.
485 // The last line gets a "*/" postfix.
486 if (LineIndex + 1 == Lines.size()) {
488 // We never need a decoration when breaking just the trailing "*/" postfix.
489 // Note that checking that Length == 0 is not enough, since Length could
490 // also be StringRef::npos.
491 if (Content[LineIndex].substr(Offset, StringRef::npos).empty()) {
492 LineLength -= Decoration.size();
498 unsigned BreakableBlockComment::getRemainingLength(unsigned LineIndex,
500 unsigned StartColumn) const {
501 return UnbreakableTailLength +
502 getRangeLength(LineIndex, Offset, StringRef::npos, StartColumn);
505 unsigned BreakableBlockComment::getContentStartColumn(unsigned LineIndex,
508 return IndentAtLineBreak;
509 return std::max(0, ContentColumn[LineIndex]);
512 void BreakableBlockComment::insertBreak(unsigned LineIndex, unsigned TailOffset,
514 WhitespaceManager &Whitespaces) const {
515 StringRef Text = Content[LineIndex].substr(TailOffset);
516 StringRef Prefix = Decoration;
517 // We need this to account for the case when we have a decoration "* " for all
518 // the lines except for the last one, where the star in "*/" acts as a
520 unsigned LocalIndentAtLineBreak = IndentAtLineBreak;
521 if (LineIndex + 1 == Lines.size() &&
522 Text.size() == Split.first + Split.second) {
523 // For the last line we need to break before "*/", but not to add "* ".
525 if (LocalIndentAtLineBreak >= 2)
526 LocalIndentAtLineBreak -= 2;
528 // The split offset is from the beginning of the line. Convert it to an offset
529 // from the beginning of the token text.
530 unsigned BreakOffsetInToken =
531 Text.data() - tokenAt(LineIndex).TokenText.data() + Split.first;
532 unsigned CharsToRemove = Split.second;
533 assert(LocalIndentAtLineBreak >= Prefix.size());
534 Whitespaces.replaceWhitespaceInToken(
535 tokenAt(LineIndex), BreakOffsetInToken, CharsToRemove, "", Prefix,
536 InPPDirective, /*Newlines=*/1,
537 /*Spaces=*/LocalIndentAtLineBreak - Prefix.size());
540 BreakableToken::Split
541 BreakableBlockComment::getReflowSplit(unsigned LineIndex,
542 llvm::Regex &CommentPragmasRegex) const {
543 if (!mayReflow(LineIndex, CommentPragmasRegex))
544 return Split(StringRef::npos, 0);
546 size_t Trimmed = Content[LineIndex].find_first_not_of(Blanks);
547 return Split(0, Trimmed != StringRef::npos ? Trimmed : 0);
550 bool BreakableBlockComment::introducesBreakBeforeToken() const {
551 // A break is introduced when we want delimiters on newline.
552 return DelimitersOnNewline &&
553 Lines[0].substr(1).find_first_not_of(Blanks) != StringRef::npos;
556 void BreakableBlockComment::reflow(unsigned LineIndex,
557 WhitespaceManager &Whitespaces) const {
558 StringRef TrimmedContent = Content[LineIndex].ltrim(Blanks);
559 // Here we need to reflow.
560 assert(Tokens[LineIndex - 1] == Tokens[LineIndex] &&
561 "Reflowing whitespace within a token");
562 // This is the offset of the end of the last line relative to the start of
563 // the token text in the token.
564 unsigned WhitespaceOffsetInToken = Content[LineIndex - 1].data() +
565 Content[LineIndex - 1].size() -
566 tokenAt(LineIndex).TokenText.data();
567 unsigned WhitespaceLength = TrimmedContent.data() -
568 tokenAt(LineIndex).TokenText.data() -
569 WhitespaceOffsetInToken;
570 Whitespaces.replaceWhitespaceInToken(
571 tokenAt(LineIndex), WhitespaceOffsetInToken,
572 /*ReplaceChars=*/WhitespaceLength, /*PreviousPostfix=*/"",
573 /*CurrentPrefix=*/ReflowPrefix, InPPDirective, /*Newlines=*/0,
577 void BreakableBlockComment::adaptStartOfLine(
578 unsigned LineIndex, WhitespaceManager &Whitespaces) const {
579 if (LineIndex == 0) {
580 if (DelimitersOnNewline) {
581 // Since we're breaking at index 1 below, the break position and the
582 // break length are the same.
583 size_t BreakLength = Lines[0].substr(1).find_first_not_of(Blanks);
584 if (BreakLength != StringRef::npos)
585 insertBreak(LineIndex, 0, Split(1, BreakLength), Whitespaces);
589 // Here no reflow with the previous line will happen.
590 // Fix the decoration of the line at LineIndex.
591 StringRef Prefix = Decoration;
592 if (Content[LineIndex].empty()) {
593 if (LineIndex + 1 == Lines.size()) {
594 if (!LastLineNeedsDecoration) {
595 // If the last line was empty, we don't need a prefix, as the */ will
596 // line up with the decoration (if it exists).
599 } else if (!Decoration.empty()) {
600 // For other empty lines, if we do have a decoration, adapt it to not
601 // contain a trailing whitespace.
602 Prefix = Prefix.substr(0, 1);
605 if (ContentColumn[LineIndex] == 1) {
606 // This line starts immediately after the decorating *.
607 Prefix = Prefix.substr(0, 1);
610 // This is the offset of the end of the last line relative to the start of the
611 // token text in the token.
612 unsigned WhitespaceOffsetInToken = Content[LineIndex - 1].data() +
613 Content[LineIndex - 1].size() -
614 tokenAt(LineIndex).TokenText.data();
615 unsigned WhitespaceLength = Content[LineIndex].data() -
616 tokenAt(LineIndex).TokenText.data() -
617 WhitespaceOffsetInToken;
618 Whitespaces.replaceWhitespaceInToken(
619 tokenAt(LineIndex), WhitespaceOffsetInToken, WhitespaceLength, "", Prefix,
620 InPPDirective, /*Newlines=*/1, ContentColumn[LineIndex] - Prefix.size());
623 BreakableToken::Split
624 BreakableBlockComment::getSplitAfterLastLine(unsigned TailOffset) const {
625 if (DelimitersOnNewline) {
626 // Replace the trailing whitespace of the last line with a newline.
627 // In case the last line is empty, the ending '*/' is already on its own
629 StringRef Line = Content.back().substr(TailOffset);
630 StringRef TrimmedLine = Line.rtrim(Blanks);
631 if (!TrimmedLine.empty())
632 return Split(TrimmedLine.size(), Line.size() - TrimmedLine.size());
634 return Split(StringRef::npos, 0);
637 bool BreakableBlockComment::mayReflow(unsigned LineIndex,
638 llvm::Regex &CommentPragmasRegex) const {
639 // Content[LineIndex] may exclude the indent after the '*' decoration. In that
640 // case, we compute the start of the comment pragma manually.
641 StringRef IndentContent = Content[LineIndex];
642 if (Lines[LineIndex].ltrim(Blanks).startswith("*")) {
643 IndentContent = Lines[LineIndex].ltrim(Blanks).substr(1);
645 return LineIndex > 0 && !CommentPragmasRegex.match(IndentContent) &&
646 mayReflowContent(Content[LineIndex]) && !Tok.Finalized &&
647 !switchesFormatting(tokenAt(LineIndex));
650 BreakableLineCommentSection::BreakableLineCommentSection(
651 const FormatToken &Token, unsigned StartColumn,
652 unsigned OriginalStartColumn, bool FirstInLine, bool InPPDirective,
653 encoding::Encoding Encoding, const FormatStyle &Style)
654 : BreakableComment(Token, StartColumn, InPPDirective, Encoding, Style) {
655 assert(Tok.is(TT_LineComment) &&
656 "line comment section must start with a line comment");
657 FormatToken *LineTok = nullptr;
658 for (const FormatToken *CurrentTok = &Tok;
659 CurrentTok && CurrentTok->is(TT_LineComment);
660 CurrentTok = CurrentTok->Next) {
661 LastLineTok = LineTok;
662 StringRef TokenText(CurrentTok->TokenText);
663 assert((TokenText.startswith("//") || TokenText.startswith("#")) &&
664 "unsupported line comment prefix, '//' and '#' are supported");
665 size_t FirstLineIndex = Lines.size();
666 TokenText.split(Lines, "\n");
667 Content.resize(Lines.size());
668 ContentColumn.resize(Lines.size());
669 OriginalContentColumn.resize(Lines.size());
670 Tokens.resize(Lines.size());
671 Prefix.resize(Lines.size());
672 OriginalPrefix.resize(Lines.size());
673 for (size_t i = FirstLineIndex, e = Lines.size(); i < e; ++i) {
674 Lines[i] = Lines[i].ltrim(Blanks);
675 // We need to trim the blanks in case this is not the first line in a
676 // multiline comment. Then the indent is included in Lines[i].
677 StringRef IndentPrefix =
678 getLineCommentIndentPrefix(Lines[i].ltrim(Blanks), Style);
679 assert((TokenText.startswith("//") || TokenText.startswith("#")) &&
680 "unsupported line comment prefix, '//' and '#' are supported");
681 OriginalPrefix[i] = Prefix[i] = IndentPrefix;
682 if (Lines[i].size() > Prefix[i].size() &&
683 isAlphanumeric(Lines[i][Prefix[i].size()])) {
684 if (Prefix[i] == "//")
686 else if (Prefix[i] == "///")
688 else if (Prefix[i] == "//!")
690 else if (Prefix[i] == "///<")
692 else if (Prefix[i] == "//!<")
694 else if (Prefix[i] == "#" &&
695 Style.Language == FormatStyle::LK_TextProto)
700 Content[i] = Lines[i].substr(IndentPrefix.size());
701 OriginalContentColumn[i] =
702 StartColumn + encoding::columnWidthWithTabs(OriginalPrefix[i],
704 Style.TabWidth, Encoding);
706 StartColumn + encoding::columnWidthWithTabs(Prefix[i], StartColumn,
707 Style.TabWidth, Encoding);
709 // Calculate the end of the non-whitespace text in this line.
710 size_t EndOfLine = Content[i].find_last_not_of(Blanks);
711 if (EndOfLine == StringRef::npos)
712 EndOfLine = Content[i].size();
715 Content[i] = Content[i].substr(0, EndOfLine);
717 LineTok = CurrentTok->Next;
718 if (CurrentTok->Next && !CurrentTok->Next->ContinuesLineCommentSection) {
719 // A line comment section needs to broken by a line comment that is
720 // preceded by at least two newlines. Note that we put this break here
721 // instead of breaking at a previous stage during parsing, since that
722 // would split the contents of the enum into two unwrapped lines in this
723 // example, which is undesirable:
725 // a, // comment about a
727 // // comment about b
731 // FIXME: Consider putting separate line comment sections as children to
732 // the unwrapped line instead.
739 BreakableLineCommentSection::getRangeLength(unsigned LineIndex, unsigned Offset,
740 StringRef::size_type Length,
741 unsigned StartColumn) const {
742 return encoding::columnWidthWithTabs(
743 Content[LineIndex].substr(Offset, Length), StartColumn, Style.TabWidth,
747 unsigned BreakableLineCommentSection::getContentStartColumn(unsigned LineIndex,
750 return OriginalContentColumn[LineIndex];
751 return ContentColumn[LineIndex];
754 void BreakableLineCommentSection::insertBreak(
755 unsigned LineIndex, unsigned TailOffset, Split Split,
756 WhitespaceManager &Whitespaces) const {
757 StringRef Text = Content[LineIndex].substr(TailOffset);
758 // Compute the offset of the split relative to the beginning of the token
760 unsigned BreakOffsetInToken =
761 Text.data() - tokenAt(LineIndex).TokenText.data() + Split.first;
762 unsigned CharsToRemove = Split.second;
763 // Compute the size of the new indent, including the size of the new prefix of
764 // the newly broken line.
765 unsigned IndentAtLineBreak = OriginalContentColumn[LineIndex] +
766 Prefix[LineIndex].size() -
767 OriginalPrefix[LineIndex].size();
768 assert(IndentAtLineBreak >= Prefix[LineIndex].size());
769 Whitespaces.replaceWhitespaceInToken(
770 tokenAt(LineIndex), BreakOffsetInToken, CharsToRemove, "",
771 Prefix[LineIndex], InPPDirective, /*Newlines=*/1,
772 /*Spaces=*/IndentAtLineBreak - Prefix[LineIndex].size());
775 BreakableComment::Split BreakableLineCommentSection::getReflowSplit(
776 unsigned LineIndex, llvm::Regex &CommentPragmasRegex) const {
777 if (!mayReflow(LineIndex, CommentPragmasRegex))
778 return Split(StringRef::npos, 0);
780 size_t Trimmed = Content[LineIndex].find_first_not_of(Blanks);
782 // In a line comment section each line is a separate token; thus, after a
783 // split we replace all whitespace before the current line comment token
784 // (which does not need to be included in the split), plus the start of the
785 // line up to where the content starts.
786 return Split(0, Trimmed != StringRef::npos ? Trimmed : 0);
789 void BreakableLineCommentSection::reflow(unsigned LineIndex,
790 WhitespaceManager &Whitespaces) const {
791 // Reflow happens between tokens. Replace the whitespace between the
792 // tokens by the empty string.
793 Whitespaces.replaceWhitespace(
794 *Tokens[LineIndex], /*Newlines=*/0, /*Spaces=*/0,
795 /*StartOfTokenColumn=*/StartColumn, /*InPPDirective=*/false);
796 // Replace the indent and prefix of the token with the reflow prefix.
797 unsigned WhitespaceLength =
798 Content[LineIndex].data() - tokenAt(LineIndex).TokenText.data();
799 Whitespaces.replaceWhitespaceInToken(*Tokens[LineIndex],
801 /*ReplaceChars=*/WhitespaceLength,
802 /*PreviousPostfix=*/"",
803 /*CurrentPrefix=*/ReflowPrefix,
804 /*InPPDirective=*/false,
809 void BreakableLineCommentSection::adaptStartOfLine(
810 unsigned LineIndex, WhitespaceManager &Whitespaces) const {
811 // If this is the first line of a token, we need to inform Whitespace Manager
812 // about it: either adapt the whitespace range preceding it, or mark it as an
813 // untouchable token.
814 // This happens for instance here:
817 if (LineIndex > 0 && Tokens[LineIndex] != Tokens[LineIndex - 1]) {
818 // This is the first line for the current token, but no reflow with the
819 // previous token is necessary. However, we still may need to adjust the
820 // start column. Note that ContentColumn[LineIndex] is the expected
821 // content column after a possible update to the prefix, hence the prefix
822 // length change is included.
823 unsigned LineColumn =
824 ContentColumn[LineIndex] -
825 (Content[LineIndex].data() - Lines[LineIndex].data()) +
826 (OriginalPrefix[LineIndex].size() - Prefix[LineIndex].size());
828 // We always want to create a replacement instead of adding an untouchable
829 // token, even if LineColumn is the same as the original column of the
830 // token. This is because WhitespaceManager doesn't align trailing
831 // comments if they are untouchable.
832 Whitespaces.replaceWhitespace(*Tokens[LineIndex],
834 /*Spaces=*/LineColumn,
835 /*StartOfTokenColumn=*/LineColumn,
836 /*InPPDirective=*/false);
838 if (OriginalPrefix[LineIndex] != Prefix[LineIndex]) {
839 // Adjust the prefix if necessary.
841 // Take care of the space possibly introduced after a decoration.
842 assert(Prefix[LineIndex] == (OriginalPrefix[LineIndex] + " ").str() &&
843 "Expecting a line comment prefix to differ from original by at most "
845 Whitespaces.replaceWhitespaceInToken(
846 tokenAt(LineIndex), OriginalPrefix[LineIndex].size(), 0, "", "",
847 /*InPPDirective=*/false, /*Newlines=*/0, /*Spaces=*/1);
851 void BreakableLineCommentSection::updateNextToken(LineState &State) const {
853 State.NextToken = LastLineTok->Next;
857 bool BreakableLineCommentSection::mayReflow(
858 unsigned LineIndex, llvm::Regex &CommentPragmasRegex) const {
859 // Line comments have the indent as part of the prefix, so we need to
860 // recompute the start of the line.
861 StringRef IndentContent = Content[LineIndex];
862 if (Lines[LineIndex].startswith("//")) {
863 IndentContent = Lines[LineIndex].substr(2);
865 // FIXME: Decide whether we want to reflow non-regular indents:
866 // Currently, we only reflow when the OriginalPrefix[LineIndex] matches the
867 // OriginalPrefix[LineIndex-1]. That means we don't reflow
868 // // text that protrudes
869 // // into text with different indent
870 // We do reflow in that case in block comments.
871 return LineIndex > 0 && !CommentPragmasRegex.match(IndentContent) &&
872 mayReflowContent(Content[LineIndex]) && !Tok.Finalized &&
873 !switchesFormatting(tokenAt(LineIndex)) &&
874 OriginalPrefix[LineIndex] == OriginalPrefix[LineIndex - 1];
877 } // namespace format