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 /// 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[] = {"//", "#", "##", "###",
49 ArrayRef<const char *> KnownPrefixes(KnownCStylePrefixes);
50 if (Style.Language == FormatStyle::LK_TextProto)
51 KnownPrefixes = KnownTextProtoPrefixes;
53 StringRef LongestPrefix;
54 for (StringRef KnownPrefix : KnownPrefixes) {
55 if (Comment.startswith(KnownPrefix)) {
56 size_t PrefixLength = KnownPrefix.size();
57 while (PrefixLength < Comment.size() && Comment[PrefixLength] == ' ')
59 if (PrefixLength > LongestPrefix.size())
60 LongestPrefix = Comment.substr(0, PrefixLength);
66 static BreakableToken::Split getCommentSplit(StringRef Text,
67 unsigned ContentStartColumn,
70 encoding::Encoding Encoding) {
71 LLVM_DEBUG(llvm::dbgs() << "Comment split: \"" << Text << ", " << ColumnLimit
72 << "\", Content start: " << ContentStartColumn
74 if (ColumnLimit <= ContentStartColumn + 1)
75 return BreakableToken::Split(StringRef::npos, 0);
77 unsigned MaxSplit = ColumnLimit - ContentStartColumn + 1;
78 unsigned MaxSplitBytes = 0;
80 for (unsigned NumChars = 0;
81 NumChars < MaxSplit && MaxSplitBytes < Text.size();) {
82 unsigned BytesInChar =
83 encoding::getCodePointNumBytes(Text[MaxSplitBytes], Encoding);
85 encoding::columnWidthWithTabs(Text.substr(MaxSplitBytes, BytesInChar),
86 ContentStartColumn, TabWidth, Encoding);
87 MaxSplitBytes += BytesInChar;
90 StringRef::size_type SpaceOffset = Text.find_last_of(Blanks, MaxSplitBytes);
92 // Do not split before a number followed by a dot: this would be interpreted
93 // as a numbered list, which would prevent re-flowing in subsequent passes.
94 static auto *const kNumberedListRegexp = new llvm::Regex("^[1-9][0-9]?\\.");
95 if (SpaceOffset != StringRef::npos &&
96 kNumberedListRegexp->match(Text.substr(SpaceOffset).ltrim(Blanks)))
97 SpaceOffset = Text.find_last_of(Blanks, SpaceOffset);
99 if (SpaceOffset == StringRef::npos ||
100 // Don't break at leading whitespace.
101 Text.find_last_not_of(Blanks, SpaceOffset) == StringRef::npos) {
102 // Make sure that we don't break at leading whitespace that
103 // reaches past MaxSplit.
104 StringRef::size_type FirstNonWhitespace = Text.find_first_not_of(Blanks);
105 if (FirstNonWhitespace == StringRef::npos)
106 // If the comment is only whitespace, we cannot split.
107 return BreakableToken::Split(StringRef::npos, 0);
108 SpaceOffset = Text.find_first_of(
109 Blanks, std::max<unsigned>(MaxSplitBytes, FirstNonWhitespace));
111 if (SpaceOffset != StringRef::npos && SpaceOffset != 0) {
112 StringRef BeforeCut = Text.substr(0, SpaceOffset).rtrim(Blanks);
113 StringRef AfterCut = Text.substr(SpaceOffset).ltrim(Blanks);
114 return BreakableToken::Split(BeforeCut.size(),
115 AfterCut.begin() - BeforeCut.end());
117 return BreakableToken::Split(StringRef::npos, 0);
120 static BreakableToken::Split
121 getStringSplit(StringRef Text, unsigned UsedColumns, unsigned ColumnLimit,
122 unsigned TabWidth, encoding::Encoding Encoding) {
123 // FIXME: Reduce unit test case.
125 return BreakableToken::Split(StringRef::npos, 0);
126 if (ColumnLimit <= UsedColumns)
127 return BreakableToken::Split(StringRef::npos, 0);
128 unsigned MaxSplit = ColumnLimit - UsedColumns;
129 StringRef::size_type SpaceOffset = 0;
130 StringRef::size_type SlashOffset = 0;
131 StringRef::size_type WordStartOffset = 0;
132 StringRef::size_type SplitPoint = 0;
133 for (unsigned Chars = 0;;) {
135 if (Text[0] == '\\') {
136 Advance = encoding::getEscapeSequenceLength(Text);
139 Advance = encoding::getCodePointNumBytes(Text[0], Encoding);
140 Chars += encoding::columnWidthWithTabs(
141 Text.substr(0, Advance), UsedColumns + Chars, TabWidth, Encoding);
144 if (Chars > MaxSplit || Text.size() <= Advance)
147 if (IsBlank(Text[0]))
148 SpaceOffset = SplitPoint;
150 SlashOffset = SplitPoint;
151 if (Advance == 1 && !isAlphanumeric(Text[0]))
152 WordStartOffset = SplitPoint;
154 SplitPoint += Advance;
155 Text = Text.substr(Advance);
158 if (SpaceOffset != 0)
159 return BreakableToken::Split(SpaceOffset + 1, 0);
160 if (SlashOffset != 0)
161 return BreakableToken::Split(SlashOffset + 1, 0);
162 if (WordStartOffset != 0)
163 return BreakableToken::Split(WordStartOffset + 1, 0);
165 return BreakableToken::Split(SplitPoint, 0);
166 return BreakableToken::Split(StringRef::npos, 0);
169 bool switchesFormatting(const FormatToken &Token) {
170 assert((Token.is(TT_BlockComment) || Token.is(TT_LineComment)) &&
171 "formatting regions are switched by comment tokens");
172 StringRef Content = Token.TokenText.substr(2).ltrim();
173 return Content.startswith("clang-format on") ||
174 Content.startswith("clang-format off");
178 BreakableToken::getLengthAfterCompression(unsigned RemainingTokenColumns,
180 // Example: consider the content
182 // - RemainingTokenColumns is the original number of columns, 10;
183 // - Split is (4, 2), denoting the two spaces between the two words;
185 // We compute the number of columns when the split is compressed into a single
189 // FIXME: Correctly measure the length of whitespace in Split.second so it
191 return RemainingTokenColumns + 1 - Split.second;
194 unsigned BreakableStringLiteral::getLineCount() const { return 1; }
196 unsigned BreakableStringLiteral::getRangeLength(unsigned LineIndex,
198 StringRef::size_type Length,
199 unsigned StartColumn) const {
200 llvm_unreachable("Getting the length of a part of the string literal "
201 "indicates that the code tries to reflow it.");
205 BreakableStringLiteral::getRemainingLength(unsigned LineIndex, unsigned Offset,
206 unsigned StartColumn) const {
207 return UnbreakableTailLength + Postfix.size() +
208 encoding::columnWidthWithTabs(Line.substr(Offset, StringRef::npos),
209 StartColumn, Style.TabWidth, Encoding);
212 unsigned BreakableStringLiteral::getContentStartColumn(unsigned LineIndex,
214 return StartColumn + Prefix.size();
217 BreakableStringLiteral::BreakableStringLiteral(
218 const FormatToken &Tok, unsigned StartColumn, StringRef Prefix,
219 StringRef Postfix, unsigned UnbreakableTailLength, bool InPPDirective,
220 encoding::Encoding Encoding, const FormatStyle &Style)
221 : BreakableToken(Tok, InPPDirective, Encoding, Style),
222 StartColumn(StartColumn), Prefix(Prefix), Postfix(Postfix),
223 UnbreakableTailLength(UnbreakableTailLength) {
224 assert(Tok.TokenText.startswith(Prefix) && Tok.TokenText.endswith(Postfix));
225 Line = Tok.TokenText.substr(
226 Prefix.size(), Tok.TokenText.size() - Prefix.size() - Postfix.size());
229 BreakableToken::Split BreakableStringLiteral::getSplit(
230 unsigned LineIndex, unsigned TailOffset, unsigned ColumnLimit,
231 unsigned ContentStartColumn, llvm::Regex &CommentPragmasRegex) const {
232 return getStringSplit(Line.substr(TailOffset), ContentStartColumn,
233 ColumnLimit - Postfix.size(), Style.TabWidth, Encoding);
236 void BreakableStringLiteral::insertBreak(unsigned LineIndex,
237 unsigned TailOffset, Split Split,
238 unsigned ContentIndent,
239 WhitespaceManager &Whitespaces) const {
240 Whitespaces.replaceWhitespaceInToken(
241 Tok, Prefix.size() + TailOffset + Split.first, Split.second, Postfix,
242 Prefix, InPPDirective, 1, StartColumn);
245 BreakableComment::BreakableComment(const FormatToken &Token,
246 unsigned StartColumn, bool InPPDirective,
247 encoding::Encoding Encoding,
248 const FormatStyle &Style)
249 : BreakableToken(Token, InPPDirective, Encoding, Style),
250 StartColumn(StartColumn) {}
252 unsigned BreakableComment::getLineCount() const { return Lines.size(); }
254 BreakableToken::Split
255 BreakableComment::getSplit(unsigned LineIndex, unsigned TailOffset,
256 unsigned ColumnLimit, unsigned ContentStartColumn,
257 llvm::Regex &CommentPragmasRegex) const {
258 // Don't break lines matching the comment pragmas regex.
259 if (CommentPragmasRegex.match(Content[LineIndex]))
260 return Split(StringRef::npos, 0);
261 return getCommentSplit(Content[LineIndex].substr(TailOffset),
262 ContentStartColumn, ColumnLimit, Style.TabWidth,
266 void BreakableComment::compressWhitespace(
267 unsigned LineIndex, unsigned TailOffset, Split Split,
268 WhitespaceManager &Whitespaces) const {
269 StringRef Text = Content[LineIndex].substr(TailOffset);
270 // Text is relative to the content line, but Whitespaces operates relative to
271 // the start of the corresponding token, so compute the start of the Split
272 // that needs to be compressed into a single space relative to the start of
274 unsigned BreakOffsetInToken =
275 Text.data() - tokenAt(LineIndex).TokenText.data() + Split.first;
276 unsigned CharsToRemove = Split.second;
277 Whitespaces.replaceWhitespaceInToken(
278 tokenAt(LineIndex), BreakOffsetInToken, CharsToRemove, "", "",
279 /*InPPDirective=*/false, /*Newlines=*/0, /*Spaces=*/1);
282 const FormatToken &BreakableComment::tokenAt(unsigned LineIndex) const {
283 return Tokens[LineIndex] ? *Tokens[LineIndex] : Tok;
286 static bool mayReflowContent(StringRef Content) {
287 Content = Content.trim(Blanks);
288 // Lines starting with '@' commonly have special meaning.
289 // Lines starting with '-', '-#', '+' or '*' are bulleted/numbered lists.
290 bool hasSpecialMeaningPrefix = false;
291 for (StringRef Prefix :
292 {"@", "TODO", "FIXME", "XXX", "-# ", "- ", "+ ", "* "}) {
293 if (Content.startswith(Prefix)) {
294 hasSpecialMeaningPrefix = true;
299 // Numbered lists may also start with a number followed by '.'
300 // To avoid issues if a line starts with a number which is actually the end
301 // of a previous line, we only consider numbers with up to 2 digits.
302 static auto *const kNumberedListRegexp = new llvm::Regex("^[1-9][0-9]?\\. ");
303 hasSpecialMeaningPrefix =
304 hasSpecialMeaningPrefix || kNumberedListRegexp->match(Content);
306 // Simple heuristic for what to reflow: content should contain at least two
307 // characters and either the first or second character must be
309 return Content.size() >= 2 && !hasSpecialMeaningPrefix &&
310 !Content.endswith("\\") &&
311 // Note that this is UTF-8 safe, since if isPunctuation(Content[0]) is
312 // true, then the first code point must be 1 byte long.
313 (!isPunctuation(Content[0]) || !isPunctuation(Content[1]));
316 BreakableBlockComment::BreakableBlockComment(
317 const FormatToken &Token, unsigned StartColumn,
318 unsigned OriginalStartColumn, bool FirstInLine, bool InPPDirective,
319 encoding::Encoding Encoding, const FormatStyle &Style)
320 : BreakableComment(Token, StartColumn, InPPDirective, Encoding, Style),
321 DelimitersOnNewline(false),
322 UnbreakableTailLength(Token.UnbreakableTailLength) {
323 assert(Tok.is(TT_BlockComment) &&
324 "block comment section must start with a block comment");
326 StringRef TokenText(Tok.TokenText);
327 assert(TokenText.startswith("/*") && TokenText.endswith("*/"));
328 TokenText.substr(2, TokenText.size() - 4).split(Lines, "\n");
330 int IndentDelta = StartColumn - OriginalStartColumn;
331 Content.resize(Lines.size());
332 Content[0] = Lines[0];
333 ContentColumn.resize(Lines.size());
334 // Account for the initial '/*'.
335 ContentColumn[0] = StartColumn + 2;
336 Tokens.resize(Lines.size());
337 for (size_t i = 1; i < Lines.size(); ++i)
338 adjustWhitespace(i, IndentDelta);
340 // Align decorations with the column of the star on the first line,
341 // that is one column after the start "/*".
342 DecorationColumn = StartColumn + 1;
344 // Account for comment decoration patterns like this:
349 if (Lines.size() >= 2 && Content[1].startswith("**") &&
350 static_cast<unsigned>(ContentColumn[1]) == StartColumn) {
351 DecorationColumn = StartColumn;
355 if (Lines.size() == 1 && !FirstInLine) {
356 // Comments for which FirstInLine is false can start on arbitrary column,
357 // and available horizontal space can be too small to align consecutive
358 // lines with the first one.
359 // FIXME: We could, probably, align them to current indentation level, but
360 // now we just wrap them without stars.
363 for (size_t i = 1, e = Lines.size(); i < e && !Decoration.empty(); ++i) {
364 // If the last line is empty, the closing "*/" will have a star.
365 if (i + 1 == e && Content[i].empty())
367 if (!Content[i].empty() && i + 1 != e && Decoration.startswith(Content[i]))
369 while (!Content[i].startswith(Decoration))
370 Decoration = Decoration.substr(0, Decoration.size() - 1);
373 LastLineNeedsDecoration = true;
374 IndentAtLineBreak = ContentColumn[0] + 1;
375 for (size_t i = 1, e = Lines.size(); i < e; ++i) {
376 if (Content[i].empty()) {
378 // Empty last line means that we already have a star as a part of the
379 // trailing */. We also need to preserve whitespace, so that */ is
380 // correctly indented.
381 LastLineNeedsDecoration = false;
382 // Align the star in the last '*/' with the stars on the previous lines.
383 if (e >= 2 && !Decoration.empty()) {
384 ContentColumn[i] = DecorationColumn;
386 } else if (Decoration.empty()) {
387 // For all other lines, set the start column to 0 if they're empty, so
388 // we do not insert trailing whitespace anywhere.
389 ContentColumn[i] = 0;
394 // The first line already excludes the star.
395 // The last line excludes the star if LastLineNeedsDecoration is false.
396 // For all other lines, adjust the line to exclude the star and
397 // (optionally) the first whitespace.
398 unsigned DecorationSize = Decoration.startswith(Content[i])
401 if (DecorationSize) {
402 ContentColumn[i] = DecorationColumn + DecorationSize;
404 Content[i] = Content[i].substr(DecorationSize);
405 if (!Decoration.startswith(Content[i]))
407 std::min<int>(IndentAtLineBreak, std::max(0, ContentColumn[i]));
409 IndentAtLineBreak = std::max<unsigned>(IndentAtLineBreak, Decoration.size());
411 // Detect a multiline jsdoc comment and set DelimitersOnNewline in that case.
412 if (Style.Language == FormatStyle::LK_JavaScript ||
413 Style.Language == FormatStyle::LK_Java) {
414 if ((Lines[0] == "*" || Lines[0].startswith("* ")) && Lines.size() > 1) {
415 // This is a multiline jsdoc comment.
416 DelimitersOnNewline = true;
417 } else if (Lines[0].startswith("* ") && Lines.size() == 1) {
418 // Detect a long single-line comment, like:
419 // /** long long long */
420 // Below, '2' is the width of '*/'.
423 encoding::columnWidthWithTabs(Lines[0], ContentColumn[0],
424 Style.TabWidth, Encoding) +
426 DelimitersOnNewline = EndColumn > Style.ColumnLimit;
431 llvm::dbgs() << "IndentAtLineBreak " << IndentAtLineBreak << "\n";
432 llvm::dbgs() << "DelimitersOnNewline " << DelimitersOnNewline << "\n";
433 for (size_t i = 0; i < Lines.size(); ++i) {
434 llvm::dbgs() << i << " |" << Content[i] << "| "
435 << "CC=" << ContentColumn[i] << "| "
436 << "IN=" << (Content[i].data() - Lines[i].data()) << "\n";
441 void BreakableBlockComment::adjustWhitespace(unsigned LineIndex,
443 // When in a preprocessor directive, the trailing backslash in a block comment
444 // is not needed, but can serve a purpose of uniformity with necessary escaped
445 // newlines outside the comment. In this case we remove it here before
446 // trimming the trailing whitespace. The backslash will be re-added later when
447 // inserting a line break.
448 size_t EndOfPreviousLine = Lines[LineIndex - 1].size();
449 if (InPPDirective && Lines[LineIndex - 1].endswith("\\"))
452 // Calculate the end of the non-whitespace text in the previous line.
454 Lines[LineIndex - 1].find_last_not_of(Blanks, EndOfPreviousLine);
455 if (EndOfPreviousLine == StringRef::npos)
456 EndOfPreviousLine = 0;
459 // Calculate the start of the non-whitespace text in the current line.
460 size_t StartOfLine = Lines[LineIndex].find_first_not_of(Blanks);
461 if (StartOfLine == StringRef::npos)
462 StartOfLine = Lines[LineIndex].rtrim("\r\n").size();
464 StringRef Whitespace = Lines[LineIndex].substr(0, StartOfLine);
465 // Adjust Lines to only contain relevant text.
466 size_t PreviousContentOffset =
467 Content[LineIndex - 1].data() - Lines[LineIndex - 1].data();
468 Content[LineIndex - 1] = Lines[LineIndex - 1].substr(
469 PreviousContentOffset, EndOfPreviousLine - PreviousContentOffset);
470 Content[LineIndex] = Lines[LineIndex].substr(StartOfLine);
472 // Adjust the start column uniformly across all lines.
473 ContentColumn[LineIndex] =
474 encoding::columnWidthWithTabs(Whitespace, 0, Style.TabWidth, Encoding) +
478 unsigned BreakableBlockComment::getRangeLength(unsigned LineIndex,
480 StringRef::size_type Length,
481 unsigned StartColumn) const {
482 unsigned LineLength =
483 encoding::columnWidthWithTabs(Content[LineIndex].substr(Offset, Length),
484 StartColumn, Style.TabWidth, Encoding);
485 // FIXME: This should go into getRemainingLength instead, but we currently
486 // break tests when putting it there. Investigate how to fix those tests.
487 // The last line gets a "*/" postfix.
488 if (LineIndex + 1 == Lines.size()) {
490 // We never need a decoration when breaking just the trailing "*/" postfix.
491 // Note that checking that Length == 0 is not enough, since Length could
492 // also be StringRef::npos.
493 if (Content[LineIndex].substr(Offset, StringRef::npos).empty()) {
494 LineLength -= Decoration.size();
500 unsigned BreakableBlockComment::getRemainingLength(unsigned LineIndex,
502 unsigned StartColumn) const {
503 return UnbreakableTailLength +
504 getRangeLength(LineIndex, Offset, StringRef::npos, StartColumn);
507 unsigned BreakableBlockComment::getContentStartColumn(unsigned LineIndex,
510 return IndentAtLineBreak;
511 return std::max(0, ContentColumn[LineIndex]);
514 const llvm::StringSet<>
515 BreakableBlockComment::ContentIndentingJavadocAnnotations = {
516 "@param", "@return", "@returns", "@throws", "@type", "@template",
517 "@see", "@deprecated", "@define", "@exports", "@mods", "@private",
520 unsigned BreakableBlockComment::getContentIndent(unsigned LineIndex) const {
521 if (Style.Language != FormatStyle::LK_Java &&
522 Style.Language != FormatStyle::LK_JavaScript)
524 // The content at LineIndex 0 of a comment like:
526 // is "* line 0", so we need to skip over the decoration in that case.
527 StringRef ContentWithNoDecoration = Content[LineIndex];
528 if (LineIndex == 0 && ContentWithNoDecoration.startswith("*")) {
529 ContentWithNoDecoration = ContentWithNoDecoration.substr(1).ltrim(Blanks);
531 StringRef FirstWord = ContentWithNoDecoration.substr(
532 0, ContentWithNoDecoration.find_first_of(Blanks));
533 if (ContentIndentingJavadocAnnotations.find(FirstWord) !=
534 ContentIndentingJavadocAnnotations.end())
535 return Style.ContinuationIndentWidth;
539 void BreakableBlockComment::insertBreak(unsigned LineIndex, unsigned TailOffset,
540 Split Split, unsigned ContentIndent,
541 WhitespaceManager &Whitespaces) const {
542 StringRef Text = Content[LineIndex].substr(TailOffset);
543 StringRef Prefix = Decoration;
544 // We need this to account for the case when we have a decoration "* " for all
545 // the lines except for the last one, where the star in "*/" acts as a
547 unsigned LocalIndentAtLineBreak = IndentAtLineBreak;
548 if (LineIndex + 1 == Lines.size() &&
549 Text.size() == Split.first + Split.second) {
550 // For the last line we need to break before "*/", but not to add "* ".
552 if (LocalIndentAtLineBreak >= 2)
553 LocalIndentAtLineBreak -= 2;
555 // The split offset is from the beginning of the line. Convert it to an offset
556 // from the beginning of the token text.
557 unsigned BreakOffsetInToken =
558 Text.data() - tokenAt(LineIndex).TokenText.data() + Split.first;
559 unsigned CharsToRemove = Split.second;
560 assert(LocalIndentAtLineBreak >= Prefix.size());
561 std::string PrefixWithTrailingIndent = Prefix;
562 for (unsigned I = 0; I < ContentIndent; ++I)
563 PrefixWithTrailingIndent += " ";
564 Whitespaces.replaceWhitespaceInToken(
565 tokenAt(LineIndex), BreakOffsetInToken, CharsToRemove, "",
566 PrefixWithTrailingIndent, InPPDirective, /*Newlines=*/1,
567 /*Spaces=*/LocalIndentAtLineBreak + ContentIndent -
568 PrefixWithTrailingIndent.size());
571 BreakableToken::Split
572 BreakableBlockComment::getReflowSplit(unsigned LineIndex,
573 llvm::Regex &CommentPragmasRegex) const {
574 if (!mayReflow(LineIndex, CommentPragmasRegex))
575 return Split(StringRef::npos, 0);
577 // If we're reflowing into a line with content indent, only reflow the next
578 // line if its starting whitespace matches the content indent.
579 size_t Trimmed = Content[LineIndex].find_first_not_of(Blanks);
581 unsigned PreviousContentIndent = getContentIndent(LineIndex - 1);
582 if (PreviousContentIndent && Trimmed != StringRef::npos &&
583 Trimmed != PreviousContentIndent)
584 return Split(StringRef::npos, 0);
587 return Split(0, Trimmed != StringRef::npos ? Trimmed : 0);
590 bool BreakableBlockComment::introducesBreakBeforeToken() const {
591 // A break is introduced when we want delimiters on newline.
592 return DelimitersOnNewline &&
593 Lines[0].substr(1).find_first_not_of(Blanks) != StringRef::npos;
596 void BreakableBlockComment::reflow(unsigned LineIndex,
597 WhitespaceManager &Whitespaces) const {
598 StringRef TrimmedContent = Content[LineIndex].ltrim(Blanks);
599 // Here we need to reflow.
600 assert(Tokens[LineIndex - 1] == Tokens[LineIndex] &&
601 "Reflowing whitespace within a token");
602 // This is the offset of the end of the last line relative to the start of
603 // the token text in the token.
604 unsigned WhitespaceOffsetInToken = Content[LineIndex - 1].data() +
605 Content[LineIndex - 1].size() -
606 tokenAt(LineIndex).TokenText.data();
607 unsigned WhitespaceLength = TrimmedContent.data() -
608 tokenAt(LineIndex).TokenText.data() -
609 WhitespaceOffsetInToken;
610 Whitespaces.replaceWhitespaceInToken(
611 tokenAt(LineIndex), WhitespaceOffsetInToken,
612 /*ReplaceChars=*/WhitespaceLength, /*PreviousPostfix=*/"",
613 /*CurrentPrefix=*/ReflowPrefix, InPPDirective, /*Newlines=*/0,
617 void BreakableBlockComment::adaptStartOfLine(
618 unsigned LineIndex, WhitespaceManager &Whitespaces) const {
619 if (LineIndex == 0) {
620 if (DelimitersOnNewline) {
621 // Since we're breaking at index 1 below, the break position and the
622 // break length are the same.
623 size_t BreakLength = Lines[0].substr(1).find_first_not_of(Blanks);
624 if (BreakLength != StringRef::npos)
625 insertBreak(LineIndex, 0, Split(1, BreakLength), /*ContentIndent=*/0,
630 // Here no reflow with the previous line will happen.
631 // Fix the decoration of the line at LineIndex.
632 StringRef Prefix = Decoration;
633 if (Content[LineIndex].empty()) {
634 if (LineIndex + 1 == Lines.size()) {
635 if (!LastLineNeedsDecoration) {
636 // If the last line was empty, we don't need a prefix, as the */ will
637 // line up with the decoration (if it exists).
640 } else if (!Decoration.empty()) {
641 // For other empty lines, if we do have a decoration, adapt it to not
642 // contain a trailing whitespace.
643 Prefix = Prefix.substr(0, 1);
646 if (ContentColumn[LineIndex] == 1) {
647 // This line starts immediately after the decorating *.
648 Prefix = Prefix.substr(0, 1);
651 // This is the offset of the end of the last line relative to the start of the
652 // token text in the token.
653 unsigned WhitespaceOffsetInToken = Content[LineIndex - 1].data() +
654 Content[LineIndex - 1].size() -
655 tokenAt(LineIndex).TokenText.data();
656 unsigned WhitespaceLength = Content[LineIndex].data() -
657 tokenAt(LineIndex).TokenText.data() -
658 WhitespaceOffsetInToken;
659 Whitespaces.replaceWhitespaceInToken(
660 tokenAt(LineIndex), WhitespaceOffsetInToken, WhitespaceLength, "", Prefix,
661 InPPDirective, /*Newlines=*/1, ContentColumn[LineIndex] - Prefix.size());
664 BreakableToken::Split
665 BreakableBlockComment::getSplitAfterLastLine(unsigned TailOffset) const {
666 if (DelimitersOnNewline) {
667 // Replace the trailing whitespace of the last line with a newline.
668 // In case the last line is empty, the ending '*/' is already on its own
670 StringRef Line = Content.back().substr(TailOffset);
671 StringRef TrimmedLine = Line.rtrim(Blanks);
672 if (!TrimmedLine.empty())
673 return Split(TrimmedLine.size(), Line.size() - TrimmedLine.size());
675 return Split(StringRef::npos, 0);
678 bool BreakableBlockComment::mayReflow(unsigned LineIndex,
679 llvm::Regex &CommentPragmasRegex) const {
680 // Content[LineIndex] may exclude the indent after the '*' decoration. In that
681 // case, we compute the start of the comment pragma manually.
682 StringRef IndentContent = Content[LineIndex];
683 if (Lines[LineIndex].ltrim(Blanks).startswith("*")) {
684 IndentContent = Lines[LineIndex].ltrim(Blanks).substr(1);
686 return LineIndex > 0 && !CommentPragmasRegex.match(IndentContent) &&
687 mayReflowContent(Content[LineIndex]) && !Tok.Finalized &&
688 !switchesFormatting(tokenAt(LineIndex));
691 BreakableLineCommentSection::BreakableLineCommentSection(
692 const FormatToken &Token, unsigned StartColumn,
693 unsigned OriginalStartColumn, bool FirstInLine, bool InPPDirective,
694 encoding::Encoding Encoding, const FormatStyle &Style)
695 : BreakableComment(Token, StartColumn, InPPDirective, Encoding, Style) {
696 assert(Tok.is(TT_LineComment) &&
697 "line comment section must start with a line comment");
698 FormatToken *LineTok = nullptr;
699 for (const FormatToken *CurrentTok = &Tok;
700 CurrentTok && CurrentTok->is(TT_LineComment);
701 CurrentTok = CurrentTok->Next) {
702 LastLineTok = LineTok;
703 StringRef TokenText(CurrentTok->TokenText);
704 assert((TokenText.startswith("//") || TokenText.startswith("#")) &&
705 "unsupported line comment prefix, '//' and '#' are supported");
706 size_t FirstLineIndex = Lines.size();
707 TokenText.split(Lines, "\n");
708 Content.resize(Lines.size());
709 ContentColumn.resize(Lines.size());
710 OriginalContentColumn.resize(Lines.size());
711 Tokens.resize(Lines.size());
712 Prefix.resize(Lines.size());
713 OriginalPrefix.resize(Lines.size());
714 for (size_t i = FirstLineIndex, e = Lines.size(); i < e; ++i) {
715 Lines[i] = Lines[i].ltrim(Blanks);
716 // We need to trim the blanks in case this is not the first line in a
717 // multiline comment. Then the indent is included in Lines[i].
718 StringRef IndentPrefix =
719 getLineCommentIndentPrefix(Lines[i].ltrim(Blanks), Style);
720 assert((TokenText.startswith("//") || TokenText.startswith("#")) &&
721 "unsupported line comment prefix, '//' and '#' are supported");
722 OriginalPrefix[i] = Prefix[i] = IndentPrefix;
723 if (Lines[i].size() > Prefix[i].size() &&
724 isAlphanumeric(Lines[i][Prefix[i].size()])) {
725 if (Prefix[i] == "//")
727 else if (Prefix[i] == "///")
729 else if (Prefix[i] == "//!")
731 else if (Prefix[i] == "///<")
733 else if (Prefix[i] == "//!<")
735 else if (Prefix[i] == "#" &&
736 Style.Language == FormatStyle::LK_TextProto)
741 Content[i] = Lines[i].substr(IndentPrefix.size());
742 OriginalContentColumn[i] =
743 StartColumn + encoding::columnWidthWithTabs(OriginalPrefix[i],
745 Style.TabWidth, Encoding);
747 StartColumn + encoding::columnWidthWithTabs(Prefix[i], StartColumn,
748 Style.TabWidth, Encoding);
750 // Calculate the end of the non-whitespace text in this line.
751 size_t EndOfLine = Content[i].find_last_not_of(Blanks);
752 if (EndOfLine == StringRef::npos)
753 EndOfLine = Content[i].size();
756 Content[i] = Content[i].substr(0, EndOfLine);
758 LineTok = CurrentTok->Next;
759 if (CurrentTok->Next && !CurrentTok->Next->ContinuesLineCommentSection) {
760 // A line comment section needs to broken by a line comment that is
761 // preceded by at least two newlines. Note that we put this break here
762 // instead of breaking at a previous stage during parsing, since that
763 // would split the contents of the enum into two unwrapped lines in this
764 // example, which is undesirable:
766 // a, // comment about a
768 // // comment about b
772 // FIXME: Consider putting separate line comment sections as children to
773 // the unwrapped line instead.
780 BreakableLineCommentSection::getRangeLength(unsigned LineIndex, unsigned Offset,
781 StringRef::size_type Length,
782 unsigned StartColumn) const {
783 return encoding::columnWidthWithTabs(
784 Content[LineIndex].substr(Offset, Length), StartColumn, Style.TabWidth,
788 unsigned BreakableLineCommentSection::getContentStartColumn(unsigned LineIndex,
791 return OriginalContentColumn[LineIndex];
792 return ContentColumn[LineIndex];
795 void BreakableLineCommentSection::insertBreak(
796 unsigned LineIndex, unsigned TailOffset, Split Split,
797 unsigned ContentIndent, WhitespaceManager &Whitespaces) const {
798 StringRef Text = Content[LineIndex].substr(TailOffset);
799 // Compute the offset of the split relative to the beginning of the token
801 unsigned BreakOffsetInToken =
802 Text.data() - tokenAt(LineIndex).TokenText.data() + Split.first;
803 unsigned CharsToRemove = Split.second;
804 // Compute the size of the new indent, including the size of the new prefix of
805 // the newly broken line.
806 unsigned IndentAtLineBreak = OriginalContentColumn[LineIndex] +
807 Prefix[LineIndex].size() -
808 OriginalPrefix[LineIndex].size();
809 assert(IndentAtLineBreak >= Prefix[LineIndex].size());
810 Whitespaces.replaceWhitespaceInToken(
811 tokenAt(LineIndex), BreakOffsetInToken, CharsToRemove, "",
812 Prefix[LineIndex], InPPDirective, /*Newlines=*/1,
813 /*Spaces=*/IndentAtLineBreak - Prefix[LineIndex].size());
816 BreakableComment::Split BreakableLineCommentSection::getReflowSplit(
817 unsigned LineIndex, llvm::Regex &CommentPragmasRegex) const {
818 if (!mayReflow(LineIndex, CommentPragmasRegex))
819 return Split(StringRef::npos, 0);
821 size_t Trimmed = Content[LineIndex].find_first_not_of(Blanks);
823 // In a line comment section each line is a separate token; thus, after a
824 // split we replace all whitespace before the current line comment token
825 // (which does not need to be included in the split), plus the start of the
826 // line up to where the content starts.
827 return Split(0, Trimmed != StringRef::npos ? Trimmed : 0);
830 void BreakableLineCommentSection::reflow(unsigned LineIndex,
831 WhitespaceManager &Whitespaces) const {
832 if (LineIndex > 0 && Tokens[LineIndex] != Tokens[LineIndex - 1]) {
833 // Reflow happens between tokens. Replace the whitespace between the
834 // tokens by the empty string.
835 Whitespaces.replaceWhitespace(
836 *Tokens[LineIndex], /*Newlines=*/0, /*Spaces=*/0,
837 /*StartOfTokenColumn=*/StartColumn, /*InPPDirective=*/false);
838 } else if (LineIndex > 0) {
839 // In case we're reflowing after the '\' in:
844 // the reflow happens inside the single comment token (it is a single line
845 // comment with an unescaped newline).
846 // Replace the whitespace between the '\' and '//' with the empty string.
848 // Offset points to after the '\' relative to start of the token.
849 unsigned Offset = Lines[LineIndex - 1].data() +
850 Lines[LineIndex - 1].size() -
851 tokenAt(LineIndex - 1).TokenText.data();
852 // WhitespaceLength is the number of chars between the '\' and the '//' on
854 unsigned WhitespaceLength =
855 Lines[LineIndex].data() - tokenAt(LineIndex).TokenText.data() - Offset;
856 Whitespaces.replaceWhitespaceInToken(*Tokens[LineIndex],
858 /*ReplaceChars=*/WhitespaceLength,
859 /*PreviousPostfix=*/"",
860 /*CurrentPrefix=*/"",
861 /*InPPDirective=*/false,
866 // Replace the indent and prefix of the token with the reflow prefix.
868 Lines[LineIndex].data() - tokenAt(LineIndex).TokenText.data();
869 unsigned WhitespaceLength =
870 Content[LineIndex].data() - Lines[LineIndex].data();
871 Whitespaces.replaceWhitespaceInToken(*Tokens[LineIndex],
873 /*ReplaceChars=*/WhitespaceLength,
874 /*PreviousPostfix=*/"",
875 /*CurrentPrefix=*/ReflowPrefix,
876 /*InPPDirective=*/false,
881 void BreakableLineCommentSection::adaptStartOfLine(
882 unsigned LineIndex, WhitespaceManager &Whitespaces) const {
883 // If this is the first line of a token, we need to inform Whitespace Manager
884 // about it: either adapt the whitespace range preceding it, or mark it as an
885 // untouchable token.
886 // This happens for instance here:
889 if (LineIndex > 0 && Tokens[LineIndex] != Tokens[LineIndex - 1]) {
890 // This is the first line for the current token, but no reflow with the
891 // previous token is necessary. However, we still may need to adjust the
892 // start column. Note that ContentColumn[LineIndex] is the expected
893 // content column after a possible update to the prefix, hence the prefix
894 // length change is included.
895 unsigned LineColumn =
896 ContentColumn[LineIndex] -
897 (Content[LineIndex].data() - Lines[LineIndex].data()) +
898 (OriginalPrefix[LineIndex].size() - Prefix[LineIndex].size());
900 // We always want to create a replacement instead of adding an untouchable
901 // token, even if LineColumn is the same as the original column of the
902 // token. This is because WhitespaceManager doesn't align trailing
903 // comments if they are untouchable.
904 Whitespaces.replaceWhitespace(*Tokens[LineIndex],
906 /*Spaces=*/LineColumn,
907 /*StartOfTokenColumn=*/LineColumn,
908 /*InPPDirective=*/false);
910 if (OriginalPrefix[LineIndex] != Prefix[LineIndex]) {
911 // Adjust the prefix if necessary.
913 // Take care of the space possibly introduced after a decoration.
914 assert(Prefix[LineIndex] == (OriginalPrefix[LineIndex] + " ").str() &&
915 "Expecting a line comment prefix to differ from original by at most "
917 Whitespaces.replaceWhitespaceInToken(
918 tokenAt(LineIndex), OriginalPrefix[LineIndex].size(), 0, "", "",
919 /*InPPDirective=*/false, /*Newlines=*/0, /*Spaces=*/1);
923 void BreakableLineCommentSection::updateNextToken(LineState &State) const {
925 State.NextToken = LastLineTok->Next;
929 bool BreakableLineCommentSection::mayReflow(
930 unsigned LineIndex, llvm::Regex &CommentPragmasRegex) const {
931 // Line comments have the indent as part of the prefix, so we need to
932 // recompute the start of the line.
933 StringRef IndentContent = Content[LineIndex];
934 if (Lines[LineIndex].startswith("//")) {
935 IndentContent = Lines[LineIndex].substr(2);
937 // FIXME: Decide whether we want to reflow non-regular indents:
938 // Currently, we only reflow when the OriginalPrefix[LineIndex] matches the
939 // OriginalPrefix[LineIndex-1]. That means we don't reflow
940 // // text that protrudes
941 // // into text with different indent
942 // We do reflow in that case in block comments.
943 return LineIndex > 0 && !CommentPragmasRegex.match(IndentContent) &&
944 mayReflowContent(Content[LineIndex]) && !Tok.Finalized &&
945 !switchesFormatting(tokenAt(LineIndex)) &&
946 OriginalPrefix[LineIndex] == OriginalPrefix[LineIndex - 1];
949 } // namespace format