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 const FormatStyle &Style) {
72 LLVM_DEBUG(llvm::dbgs() << "Comment split: \"" << Text
73 << "\", Column limit: " << ColumnLimit
74 << ", Content start: " << ContentStartColumn << "\n");
75 if (ColumnLimit <= ContentStartColumn + 1)
76 return BreakableToken::Split(StringRef::npos, 0);
78 unsigned MaxSplit = ColumnLimit - ContentStartColumn + 1;
79 unsigned MaxSplitBytes = 0;
81 for (unsigned NumChars = 0;
82 NumChars < MaxSplit && MaxSplitBytes < Text.size();) {
83 unsigned BytesInChar =
84 encoding::getCodePointNumBytes(Text[MaxSplitBytes], Encoding);
86 encoding::columnWidthWithTabs(Text.substr(MaxSplitBytes, BytesInChar),
87 ContentStartColumn, TabWidth, Encoding);
88 MaxSplitBytes += BytesInChar;
91 StringRef::size_type SpaceOffset = Text.find_last_of(Blanks, MaxSplitBytes);
93 static auto *const kNumberedListRegexp = new llvm::Regex("^[1-9][0-9]?\\.");
94 while (SpaceOffset != StringRef::npos) {
95 // Do not split before a number followed by a dot: this would be interpreted
96 // as a numbered list, which would prevent re-flowing in subsequent passes.
97 if (kNumberedListRegexp->match(Text.substr(SpaceOffset).ltrim(Blanks)))
98 SpaceOffset = Text.find_last_of(Blanks, SpaceOffset);
99 // In JavaScript, some @tags can be followed by {, and machinery that parses
100 // these comments will fail to understand the comment if followed by a line
101 // break. So avoid ever breaking before a {.
102 else if (Style.Language == FormatStyle::LK_JavaScript &&
103 SpaceOffset + 1 < Text.size() && Text[SpaceOffset + 1] == '{')
104 SpaceOffset = Text.find_last_of(Blanks, SpaceOffset);
109 if (SpaceOffset == StringRef::npos ||
110 // Don't break at leading whitespace.
111 Text.find_last_not_of(Blanks, SpaceOffset) == StringRef::npos) {
112 // Make sure that we don't break at leading whitespace that
113 // reaches past MaxSplit.
114 StringRef::size_type FirstNonWhitespace = Text.find_first_not_of(Blanks);
115 if (FirstNonWhitespace == StringRef::npos)
116 // If the comment is only whitespace, we cannot split.
117 return BreakableToken::Split(StringRef::npos, 0);
118 SpaceOffset = Text.find_first_of(
119 Blanks, std::max<unsigned>(MaxSplitBytes, FirstNonWhitespace));
121 if (SpaceOffset != StringRef::npos && SpaceOffset != 0) {
122 // adaptStartOfLine will break after lines starting with /** if the comment
123 // is broken anywhere. Avoid emitting this break twice here.
124 // Example: in /** longtextcomesherethatbreaks */ (with ColumnLimit 20) will
125 // insert a break after /**, so this code must not insert the same break.
126 if (SpaceOffset == 1 && Text[SpaceOffset - 1] == '*')
127 return BreakableToken::Split(StringRef::npos, 0);
128 StringRef BeforeCut = Text.substr(0, SpaceOffset).rtrim(Blanks);
129 StringRef AfterCut = Text.substr(SpaceOffset).ltrim(Blanks);
130 return BreakableToken::Split(BeforeCut.size(),
131 AfterCut.begin() - BeforeCut.end());
133 return BreakableToken::Split(StringRef::npos, 0);
136 static BreakableToken::Split
137 getStringSplit(StringRef Text, unsigned UsedColumns, unsigned ColumnLimit,
138 unsigned TabWidth, encoding::Encoding Encoding) {
139 // FIXME: Reduce unit test case.
141 return BreakableToken::Split(StringRef::npos, 0);
142 if (ColumnLimit <= UsedColumns)
143 return BreakableToken::Split(StringRef::npos, 0);
144 unsigned MaxSplit = ColumnLimit - UsedColumns;
145 StringRef::size_type SpaceOffset = 0;
146 StringRef::size_type SlashOffset = 0;
147 StringRef::size_type WordStartOffset = 0;
148 StringRef::size_type SplitPoint = 0;
149 for (unsigned Chars = 0;;) {
151 if (Text[0] == '\\') {
152 Advance = encoding::getEscapeSequenceLength(Text);
155 Advance = encoding::getCodePointNumBytes(Text[0], Encoding);
156 Chars += encoding::columnWidthWithTabs(
157 Text.substr(0, Advance), UsedColumns + Chars, TabWidth, Encoding);
160 if (Chars > MaxSplit || Text.size() <= Advance)
163 if (IsBlank(Text[0]))
164 SpaceOffset = SplitPoint;
166 SlashOffset = SplitPoint;
167 if (Advance == 1 && !isAlphanumeric(Text[0]))
168 WordStartOffset = SplitPoint;
170 SplitPoint += Advance;
171 Text = Text.substr(Advance);
174 if (SpaceOffset != 0)
175 return BreakableToken::Split(SpaceOffset + 1, 0);
176 if (SlashOffset != 0)
177 return BreakableToken::Split(SlashOffset + 1, 0);
178 if (WordStartOffset != 0)
179 return BreakableToken::Split(WordStartOffset + 1, 0);
181 return BreakableToken::Split(SplitPoint, 0);
182 return BreakableToken::Split(StringRef::npos, 0);
185 bool switchesFormatting(const FormatToken &Token) {
186 assert((Token.is(TT_BlockComment) || Token.is(TT_LineComment)) &&
187 "formatting regions are switched by comment tokens");
188 StringRef Content = Token.TokenText.substr(2).ltrim();
189 return Content.startswith("clang-format on") ||
190 Content.startswith("clang-format off");
194 BreakableToken::getLengthAfterCompression(unsigned RemainingTokenColumns,
196 // Example: consider the content
198 // - RemainingTokenColumns is the original number of columns, 10;
199 // - Split is (4, 2), denoting the two spaces between the two words;
201 // We compute the number of columns when the split is compressed into a single
205 // FIXME: Correctly measure the length of whitespace in Split.second so it
207 return RemainingTokenColumns + 1 - Split.second;
210 unsigned BreakableStringLiteral::getLineCount() const { return 1; }
212 unsigned BreakableStringLiteral::getRangeLength(unsigned LineIndex,
214 StringRef::size_type Length,
215 unsigned StartColumn) const {
216 llvm_unreachable("Getting the length of a part of the string literal "
217 "indicates that the code tries to reflow it.");
221 BreakableStringLiteral::getRemainingLength(unsigned LineIndex, unsigned Offset,
222 unsigned StartColumn) const {
223 return UnbreakableTailLength + Postfix.size() +
224 encoding::columnWidthWithTabs(Line.substr(Offset, StringRef::npos),
225 StartColumn, Style.TabWidth, Encoding);
228 unsigned BreakableStringLiteral::getContentStartColumn(unsigned LineIndex,
230 return StartColumn + Prefix.size();
233 BreakableStringLiteral::BreakableStringLiteral(
234 const FormatToken &Tok, unsigned StartColumn, StringRef Prefix,
235 StringRef Postfix, unsigned UnbreakableTailLength, bool InPPDirective,
236 encoding::Encoding Encoding, const FormatStyle &Style)
237 : BreakableToken(Tok, InPPDirective, Encoding, Style),
238 StartColumn(StartColumn), Prefix(Prefix), Postfix(Postfix),
239 UnbreakableTailLength(UnbreakableTailLength) {
240 assert(Tok.TokenText.startswith(Prefix) && Tok.TokenText.endswith(Postfix));
241 Line = Tok.TokenText.substr(
242 Prefix.size(), Tok.TokenText.size() - Prefix.size() - Postfix.size());
245 BreakableToken::Split BreakableStringLiteral::getSplit(
246 unsigned LineIndex, unsigned TailOffset, unsigned ColumnLimit,
247 unsigned ContentStartColumn, llvm::Regex &CommentPragmasRegex) const {
248 return getStringSplit(Line.substr(TailOffset), ContentStartColumn,
249 ColumnLimit - Postfix.size(), Style.TabWidth, Encoding);
252 void BreakableStringLiteral::insertBreak(unsigned LineIndex,
253 unsigned TailOffset, Split Split,
254 unsigned ContentIndent,
255 WhitespaceManager &Whitespaces) const {
256 Whitespaces.replaceWhitespaceInToken(
257 Tok, Prefix.size() + TailOffset + Split.first, Split.second, Postfix,
258 Prefix, InPPDirective, 1, StartColumn);
261 BreakableComment::BreakableComment(const FormatToken &Token,
262 unsigned StartColumn, bool InPPDirective,
263 encoding::Encoding Encoding,
264 const FormatStyle &Style)
265 : BreakableToken(Token, InPPDirective, Encoding, Style),
266 StartColumn(StartColumn) {}
268 unsigned BreakableComment::getLineCount() const { return Lines.size(); }
270 BreakableToken::Split
271 BreakableComment::getSplit(unsigned LineIndex, unsigned TailOffset,
272 unsigned ColumnLimit, unsigned ContentStartColumn,
273 llvm::Regex &CommentPragmasRegex) const {
274 // Don't break lines matching the comment pragmas regex.
275 if (CommentPragmasRegex.match(Content[LineIndex]))
276 return Split(StringRef::npos, 0);
277 return getCommentSplit(Content[LineIndex].substr(TailOffset),
278 ContentStartColumn, ColumnLimit, Style.TabWidth,
282 void BreakableComment::compressWhitespace(
283 unsigned LineIndex, unsigned TailOffset, Split Split,
284 WhitespaceManager &Whitespaces) const {
285 StringRef Text = Content[LineIndex].substr(TailOffset);
286 // Text is relative to the content line, but Whitespaces operates relative to
287 // the start of the corresponding token, so compute the start of the Split
288 // that needs to be compressed into a single space relative to the start of
290 unsigned BreakOffsetInToken =
291 Text.data() - tokenAt(LineIndex).TokenText.data() + Split.first;
292 unsigned CharsToRemove = Split.second;
293 Whitespaces.replaceWhitespaceInToken(
294 tokenAt(LineIndex), BreakOffsetInToken, CharsToRemove, "", "",
295 /*InPPDirective=*/false, /*Newlines=*/0, /*Spaces=*/1);
298 const FormatToken &BreakableComment::tokenAt(unsigned LineIndex) const {
299 return Tokens[LineIndex] ? *Tokens[LineIndex] : Tok;
302 static bool mayReflowContent(StringRef Content) {
303 Content = Content.trim(Blanks);
304 // Lines starting with '@' commonly have special meaning.
305 // Lines starting with '-', '-#', '+' or '*' are bulleted/numbered lists.
306 bool hasSpecialMeaningPrefix = false;
307 for (StringRef Prefix :
308 {"@", "TODO", "FIXME", "XXX", "-# ", "- ", "+ ", "* "}) {
309 if (Content.startswith(Prefix)) {
310 hasSpecialMeaningPrefix = true;
315 // Numbered lists may also start with a number followed by '.'
316 // To avoid issues if a line starts with a number which is actually the end
317 // of a previous line, we only consider numbers with up to 2 digits.
318 static auto *const kNumberedListRegexp = new llvm::Regex("^[1-9][0-9]?\\. ");
319 hasSpecialMeaningPrefix =
320 hasSpecialMeaningPrefix || kNumberedListRegexp->match(Content);
322 // Simple heuristic for what to reflow: content should contain at least two
323 // characters and either the first or second character must be
325 return Content.size() >= 2 && !hasSpecialMeaningPrefix &&
326 !Content.endswith("\\") &&
327 // Note that this is UTF-8 safe, since if isPunctuation(Content[0]) is
328 // true, then the first code point must be 1 byte long.
329 (!isPunctuation(Content[0]) || !isPunctuation(Content[1]));
332 BreakableBlockComment::BreakableBlockComment(
333 const FormatToken &Token, unsigned StartColumn,
334 unsigned OriginalStartColumn, bool FirstInLine, bool InPPDirective,
335 encoding::Encoding Encoding, const FormatStyle &Style)
336 : BreakableComment(Token, StartColumn, InPPDirective, Encoding, Style),
337 DelimitersOnNewline(false),
338 UnbreakableTailLength(Token.UnbreakableTailLength) {
339 assert(Tok.is(TT_BlockComment) &&
340 "block comment section must start with a block comment");
342 StringRef TokenText(Tok.TokenText);
343 assert(TokenText.startswith("/*") && TokenText.endswith("*/"));
344 TokenText.substr(2, TokenText.size() - 4).split(Lines, "\n");
346 int IndentDelta = StartColumn - OriginalStartColumn;
347 Content.resize(Lines.size());
348 Content[0] = Lines[0];
349 ContentColumn.resize(Lines.size());
350 // Account for the initial '/*'.
351 ContentColumn[0] = StartColumn + 2;
352 Tokens.resize(Lines.size());
353 for (size_t i = 1; i < Lines.size(); ++i)
354 adjustWhitespace(i, IndentDelta);
356 // Align decorations with the column of the star on the first line,
357 // that is one column after the start "/*".
358 DecorationColumn = StartColumn + 1;
360 // Account for comment decoration patterns like this:
365 if (Lines.size() >= 2 && Content[1].startswith("**") &&
366 static_cast<unsigned>(ContentColumn[1]) == StartColumn) {
367 DecorationColumn = StartColumn;
371 if (Lines.size() == 1 && !FirstInLine) {
372 // Comments for which FirstInLine is false can start on arbitrary column,
373 // and available horizontal space can be too small to align consecutive
374 // lines with the first one.
375 // FIXME: We could, probably, align them to current indentation level, but
376 // now we just wrap them without stars.
379 for (size_t i = 1, e = Lines.size(); i < e && !Decoration.empty(); ++i) {
380 // If the last line is empty, the closing "*/" will have a star.
381 if (i + 1 == e && Content[i].empty())
383 if (!Content[i].empty() && i + 1 != e && Decoration.startswith(Content[i]))
385 while (!Content[i].startswith(Decoration))
386 Decoration = Decoration.substr(0, Decoration.size() - 1);
389 LastLineNeedsDecoration = true;
390 IndentAtLineBreak = ContentColumn[0] + 1;
391 for (size_t i = 1, e = Lines.size(); i < e; ++i) {
392 if (Content[i].empty()) {
394 // Empty last line means that we already have a star as a part of the
395 // trailing */. We also need to preserve whitespace, so that */ is
396 // correctly indented.
397 LastLineNeedsDecoration = false;
398 // Align the star in the last '*/' with the stars on the previous lines.
399 if (e >= 2 && !Decoration.empty()) {
400 ContentColumn[i] = DecorationColumn;
402 } else if (Decoration.empty()) {
403 // For all other lines, set the start column to 0 if they're empty, so
404 // we do not insert trailing whitespace anywhere.
405 ContentColumn[i] = 0;
410 // The first line already excludes the star.
411 // The last line excludes the star if LastLineNeedsDecoration is false.
412 // For all other lines, adjust the line to exclude the star and
413 // (optionally) the first whitespace.
414 unsigned DecorationSize = Decoration.startswith(Content[i])
417 if (DecorationSize) {
418 ContentColumn[i] = DecorationColumn + DecorationSize;
420 Content[i] = Content[i].substr(DecorationSize);
421 if (!Decoration.startswith(Content[i]))
423 std::min<int>(IndentAtLineBreak, std::max(0, ContentColumn[i]));
425 IndentAtLineBreak = std::max<unsigned>(IndentAtLineBreak, Decoration.size());
427 // Detect a multiline jsdoc comment and set DelimitersOnNewline in that case.
428 if (Style.Language == FormatStyle::LK_JavaScript ||
429 Style.Language == FormatStyle::LK_Java) {
430 if ((Lines[0] == "*" || Lines[0].startswith("* ")) && Lines.size() > 1) {
431 // This is a multiline jsdoc comment.
432 DelimitersOnNewline = true;
433 } else if (Lines[0].startswith("* ") && Lines.size() == 1) {
434 // Detect a long single-line comment, like:
435 // /** long long long */
436 // Below, '2' is the width of '*/'.
439 encoding::columnWidthWithTabs(Lines[0], ContentColumn[0],
440 Style.TabWidth, Encoding) +
442 DelimitersOnNewline = EndColumn > Style.ColumnLimit;
447 llvm::dbgs() << "IndentAtLineBreak " << IndentAtLineBreak << "\n";
448 llvm::dbgs() << "DelimitersOnNewline " << DelimitersOnNewline << "\n";
449 for (size_t i = 0; i < Lines.size(); ++i) {
450 llvm::dbgs() << i << " |" << Content[i] << "| "
451 << "CC=" << ContentColumn[i] << "| "
452 << "IN=" << (Content[i].data() - Lines[i].data()) << "\n";
457 void BreakableBlockComment::adjustWhitespace(unsigned LineIndex,
459 // When in a preprocessor directive, the trailing backslash in a block comment
460 // is not needed, but can serve a purpose of uniformity with necessary escaped
461 // newlines outside the comment. In this case we remove it here before
462 // trimming the trailing whitespace. The backslash will be re-added later when
463 // inserting a line break.
464 size_t EndOfPreviousLine = Lines[LineIndex - 1].size();
465 if (InPPDirective && Lines[LineIndex - 1].endswith("\\"))
468 // Calculate the end of the non-whitespace text in the previous line.
470 Lines[LineIndex - 1].find_last_not_of(Blanks, EndOfPreviousLine);
471 if (EndOfPreviousLine == StringRef::npos)
472 EndOfPreviousLine = 0;
475 // Calculate the start of the non-whitespace text in the current line.
476 size_t StartOfLine = Lines[LineIndex].find_first_not_of(Blanks);
477 if (StartOfLine == StringRef::npos)
478 StartOfLine = Lines[LineIndex].rtrim("\r\n").size();
480 StringRef Whitespace = Lines[LineIndex].substr(0, StartOfLine);
481 // Adjust Lines to only contain relevant text.
482 size_t PreviousContentOffset =
483 Content[LineIndex - 1].data() - Lines[LineIndex - 1].data();
484 Content[LineIndex - 1] = Lines[LineIndex - 1].substr(
485 PreviousContentOffset, EndOfPreviousLine - PreviousContentOffset);
486 Content[LineIndex] = Lines[LineIndex].substr(StartOfLine);
488 // Adjust the start column uniformly across all lines.
489 ContentColumn[LineIndex] =
490 encoding::columnWidthWithTabs(Whitespace, 0, Style.TabWidth, Encoding) +
494 unsigned BreakableBlockComment::getRangeLength(unsigned LineIndex,
496 StringRef::size_type Length,
497 unsigned StartColumn) const {
498 unsigned LineLength =
499 encoding::columnWidthWithTabs(Content[LineIndex].substr(Offset, Length),
500 StartColumn, Style.TabWidth, Encoding);
501 // FIXME: This should go into getRemainingLength instead, but we currently
502 // break tests when putting it there. Investigate how to fix those tests.
503 // The last line gets a "*/" postfix.
504 if (LineIndex + 1 == Lines.size()) {
506 // We never need a decoration when breaking just the trailing "*/" postfix.
507 // Note that checking that Length == 0 is not enough, since Length could
508 // also be StringRef::npos.
509 if (Content[LineIndex].substr(Offset, StringRef::npos).empty()) {
510 LineLength -= Decoration.size();
516 unsigned BreakableBlockComment::getRemainingLength(unsigned LineIndex,
518 unsigned StartColumn) const {
519 return UnbreakableTailLength +
520 getRangeLength(LineIndex, Offset, StringRef::npos, StartColumn);
523 unsigned BreakableBlockComment::getContentStartColumn(unsigned LineIndex,
526 return IndentAtLineBreak;
527 return std::max(0, ContentColumn[LineIndex]);
530 const llvm::StringSet<>
531 BreakableBlockComment::ContentIndentingJavadocAnnotations = {
532 "@param", "@return", "@returns", "@throws", "@type", "@template",
533 "@see", "@deprecated", "@define", "@exports", "@mods", "@private",
536 unsigned BreakableBlockComment::getContentIndent(unsigned LineIndex) const {
537 if (Style.Language != FormatStyle::LK_Java &&
538 Style.Language != FormatStyle::LK_JavaScript)
540 // The content at LineIndex 0 of a comment like:
542 // is "* line 0", so we need to skip over the decoration in that case.
543 StringRef ContentWithNoDecoration = Content[LineIndex];
544 if (LineIndex == 0 && ContentWithNoDecoration.startswith("*")) {
545 ContentWithNoDecoration = ContentWithNoDecoration.substr(1).ltrim(Blanks);
547 StringRef FirstWord = ContentWithNoDecoration.substr(
548 0, ContentWithNoDecoration.find_first_of(Blanks));
549 if (ContentIndentingJavadocAnnotations.find(FirstWord) !=
550 ContentIndentingJavadocAnnotations.end())
551 return Style.ContinuationIndentWidth;
555 void BreakableBlockComment::insertBreak(unsigned LineIndex, unsigned TailOffset,
556 Split Split, unsigned ContentIndent,
557 WhitespaceManager &Whitespaces) const {
558 StringRef Text = Content[LineIndex].substr(TailOffset);
559 StringRef Prefix = Decoration;
560 // We need this to account for the case when we have a decoration "* " for all
561 // the lines except for the last one, where the star in "*/" acts as a
563 unsigned LocalIndentAtLineBreak = IndentAtLineBreak;
564 if (LineIndex + 1 == Lines.size() &&
565 Text.size() == Split.first + Split.second) {
566 // For the last line we need to break before "*/", but not to add "* ".
568 if (LocalIndentAtLineBreak >= 2)
569 LocalIndentAtLineBreak -= 2;
571 // The split offset is from the beginning of the line. Convert it to an offset
572 // from the beginning of the token text.
573 unsigned BreakOffsetInToken =
574 Text.data() - tokenAt(LineIndex).TokenText.data() + Split.first;
575 unsigned CharsToRemove = Split.second;
576 assert(LocalIndentAtLineBreak >= Prefix.size());
577 std::string PrefixWithTrailingIndent = Prefix;
578 for (unsigned I = 0; I < ContentIndent; ++I)
579 PrefixWithTrailingIndent += " ";
580 Whitespaces.replaceWhitespaceInToken(
581 tokenAt(LineIndex), BreakOffsetInToken, CharsToRemove, "",
582 PrefixWithTrailingIndent, InPPDirective, /*Newlines=*/1,
583 /*Spaces=*/LocalIndentAtLineBreak + ContentIndent -
584 PrefixWithTrailingIndent.size());
587 BreakableToken::Split
588 BreakableBlockComment::getReflowSplit(unsigned LineIndex,
589 llvm::Regex &CommentPragmasRegex) const {
590 if (!mayReflow(LineIndex, CommentPragmasRegex))
591 return Split(StringRef::npos, 0);
593 // If we're reflowing into a line with content indent, only reflow the next
594 // line if its starting whitespace matches the content indent.
595 size_t Trimmed = Content[LineIndex].find_first_not_of(Blanks);
597 unsigned PreviousContentIndent = getContentIndent(LineIndex - 1);
598 if (PreviousContentIndent && Trimmed != StringRef::npos &&
599 Trimmed != PreviousContentIndent)
600 return Split(StringRef::npos, 0);
603 return Split(0, Trimmed != StringRef::npos ? Trimmed : 0);
606 bool BreakableBlockComment::introducesBreakBeforeToken() const {
607 // A break is introduced when we want delimiters on newline.
608 return DelimitersOnNewline &&
609 Lines[0].substr(1).find_first_not_of(Blanks) != StringRef::npos;
612 void BreakableBlockComment::reflow(unsigned LineIndex,
613 WhitespaceManager &Whitespaces) const {
614 StringRef TrimmedContent = Content[LineIndex].ltrim(Blanks);
615 // Here we need to reflow.
616 assert(Tokens[LineIndex - 1] == Tokens[LineIndex] &&
617 "Reflowing whitespace within a token");
618 // This is the offset of the end of the last line relative to the start of
619 // the token text in the token.
620 unsigned WhitespaceOffsetInToken = Content[LineIndex - 1].data() +
621 Content[LineIndex - 1].size() -
622 tokenAt(LineIndex).TokenText.data();
623 unsigned WhitespaceLength = TrimmedContent.data() -
624 tokenAt(LineIndex).TokenText.data() -
625 WhitespaceOffsetInToken;
626 Whitespaces.replaceWhitespaceInToken(
627 tokenAt(LineIndex), WhitespaceOffsetInToken,
628 /*ReplaceChars=*/WhitespaceLength, /*PreviousPostfix=*/"",
629 /*CurrentPrefix=*/ReflowPrefix, InPPDirective, /*Newlines=*/0,
633 void BreakableBlockComment::adaptStartOfLine(
634 unsigned LineIndex, WhitespaceManager &Whitespaces) const {
635 if (LineIndex == 0) {
636 if (DelimitersOnNewline) {
637 // Since we're breaking at index 1 below, the break position and the
638 // break length are the same.
639 // Note: this works because getCommentSplit is careful never to split at
640 // the beginning of a line.
641 size_t BreakLength = Lines[0].substr(1).find_first_not_of(Blanks);
642 if (BreakLength != StringRef::npos)
643 insertBreak(LineIndex, 0, Split(1, BreakLength), /*ContentIndent=*/0,
648 // Here no reflow with the previous line will happen.
649 // Fix the decoration of the line at LineIndex.
650 StringRef Prefix = Decoration;
651 if (Content[LineIndex].empty()) {
652 if (LineIndex + 1 == Lines.size()) {
653 if (!LastLineNeedsDecoration) {
654 // If the last line was empty, we don't need a prefix, as the */ will
655 // line up with the decoration (if it exists).
658 } else if (!Decoration.empty()) {
659 // For other empty lines, if we do have a decoration, adapt it to not
660 // contain a trailing whitespace.
661 Prefix = Prefix.substr(0, 1);
664 if (ContentColumn[LineIndex] == 1) {
665 // This line starts immediately after the decorating *.
666 Prefix = Prefix.substr(0, 1);
669 // This is the offset of the end of the last line relative to the start of the
670 // token text in the token.
671 unsigned WhitespaceOffsetInToken = Content[LineIndex - 1].data() +
672 Content[LineIndex - 1].size() -
673 tokenAt(LineIndex).TokenText.data();
674 unsigned WhitespaceLength = Content[LineIndex].data() -
675 tokenAt(LineIndex).TokenText.data() -
676 WhitespaceOffsetInToken;
677 Whitespaces.replaceWhitespaceInToken(
678 tokenAt(LineIndex), WhitespaceOffsetInToken, WhitespaceLength, "", Prefix,
679 InPPDirective, /*Newlines=*/1, ContentColumn[LineIndex] - Prefix.size());
682 BreakableToken::Split
683 BreakableBlockComment::getSplitAfterLastLine(unsigned TailOffset) const {
684 if (DelimitersOnNewline) {
685 // Replace the trailing whitespace of the last line with a newline.
686 // In case the last line is empty, the ending '*/' is already on its own
688 StringRef Line = Content.back().substr(TailOffset);
689 StringRef TrimmedLine = Line.rtrim(Blanks);
690 if (!TrimmedLine.empty())
691 return Split(TrimmedLine.size(), Line.size() - TrimmedLine.size());
693 return Split(StringRef::npos, 0);
696 bool BreakableBlockComment::mayReflow(unsigned LineIndex,
697 llvm::Regex &CommentPragmasRegex) const {
698 // Content[LineIndex] may exclude the indent after the '*' decoration. In that
699 // case, we compute the start of the comment pragma manually.
700 StringRef IndentContent = Content[LineIndex];
701 if (Lines[LineIndex].ltrim(Blanks).startswith("*")) {
702 IndentContent = Lines[LineIndex].ltrim(Blanks).substr(1);
704 return LineIndex > 0 && !CommentPragmasRegex.match(IndentContent) &&
705 mayReflowContent(Content[LineIndex]) && !Tok.Finalized &&
706 !switchesFormatting(tokenAt(LineIndex));
709 BreakableLineCommentSection::BreakableLineCommentSection(
710 const FormatToken &Token, unsigned StartColumn,
711 unsigned OriginalStartColumn, bool FirstInLine, bool InPPDirective,
712 encoding::Encoding Encoding, const FormatStyle &Style)
713 : BreakableComment(Token, StartColumn, InPPDirective, Encoding, Style) {
714 assert(Tok.is(TT_LineComment) &&
715 "line comment section must start with a line comment");
716 FormatToken *LineTok = nullptr;
717 for (const FormatToken *CurrentTok = &Tok;
718 CurrentTok && CurrentTok->is(TT_LineComment);
719 CurrentTok = CurrentTok->Next) {
720 LastLineTok = LineTok;
721 StringRef TokenText(CurrentTok->TokenText);
722 assert((TokenText.startswith("//") || TokenText.startswith("#")) &&
723 "unsupported line comment prefix, '//' and '#' are supported");
724 size_t FirstLineIndex = Lines.size();
725 TokenText.split(Lines, "\n");
726 Content.resize(Lines.size());
727 ContentColumn.resize(Lines.size());
728 OriginalContentColumn.resize(Lines.size());
729 Tokens.resize(Lines.size());
730 Prefix.resize(Lines.size());
731 OriginalPrefix.resize(Lines.size());
732 for (size_t i = FirstLineIndex, e = Lines.size(); i < e; ++i) {
733 Lines[i] = Lines[i].ltrim(Blanks);
734 // We need to trim the blanks in case this is not the first line in a
735 // multiline comment. Then the indent is included in Lines[i].
736 StringRef IndentPrefix =
737 getLineCommentIndentPrefix(Lines[i].ltrim(Blanks), Style);
738 assert((TokenText.startswith("//") || TokenText.startswith("#")) &&
739 "unsupported line comment prefix, '//' and '#' are supported");
740 OriginalPrefix[i] = Prefix[i] = IndentPrefix;
741 if (Lines[i].size() > Prefix[i].size() &&
742 isAlphanumeric(Lines[i][Prefix[i].size()])) {
743 if (Prefix[i] == "//")
745 else if (Prefix[i] == "///")
747 else if (Prefix[i] == "//!")
749 else if (Prefix[i] == "///<")
751 else if (Prefix[i] == "//!<")
753 else if (Prefix[i] == "#" &&
754 Style.Language == FormatStyle::LK_TextProto)
759 Content[i] = Lines[i].substr(IndentPrefix.size());
760 OriginalContentColumn[i] =
761 StartColumn + encoding::columnWidthWithTabs(OriginalPrefix[i],
763 Style.TabWidth, Encoding);
765 StartColumn + encoding::columnWidthWithTabs(Prefix[i], StartColumn,
766 Style.TabWidth, Encoding);
768 // Calculate the end of the non-whitespace text in this line.
769 size_t EndOfLine = Content[i].find_last_not_of(Blanks);
770 if (EndOfLine == StringRef::npos)
771 EndOfLine = Content[i].size();
774 Content[i] = Content[i].substr(0, EndOfLine);
776 LineTok = CurrentTok->Next;
777 if (CurrentTok->Next && !CurrentTok->Next->ContinuesLineCommentSection) {
778 // A line comment section needs to broken by a line comment that is
779 // preceded by at least two newlines. Note that we put this break here
780 // instead of breaking at a previous stage during parsing, since that
781 // would split the contents of the enum into two unwrapped lines in this
782 // example, which is undesirable:
784 // a, // comment about a
786 // // comment about b
790 // FIXME: Consider putting separate line comment sections as children to
791 // the unwrapped line instead.
798 BreakableLineCommentSection::getRangeLength(unsigned LineIndex, unsigned Offset,
799 StringRef::size_type Length,
800 unsigned StartColumn) const {
801 return encoding::columnWidthWithTabs(
802 Content[LineIndex].substr(Offset, Length), StartColumn, Style.TabWidth,
806 unsigned BreakableLineCommentSection::getContentStartColumn(unsigned LineIndex,
809 return OriginalContentColumn[LineIndex];
810 return ContentColumn[LineIndex];
813 void BreakableLineCommentSection::insertBreak(
814 unsigned LineIndex, unsigned TailOffset, Split Split,
815 unsigned ContentIndent, WhitespaceManager &Whitespaces) const {
816 StringRef Text = Content[LineIndex].substr(TailOffset);
817 // Compute the offset of the split relative to the beginning of the token
819 unsigned BreakOffsetInToken =
820 Text.data() - tokenAt(LineIndex).TokenText.data() + Split.first;
821 unsigned CharsToRemove = Split.second;
822 // Compute the size of the new indent, including the size of the new prefix of
823 // the newly broken line.
824 unsigned IndentAtLineBreak = OriginalContentColumn[LineIndex] +
825 Prefix[LineIndex].size() -
826 OriginalPrefix[LineIndex].size();
827 assert(IndentAtLineBreak >= Prefix[LineIndex].size());
828 Whitespaces.replaceWhitespaceInToken(
829 tokenAt(LineIndex), BreakOffsetInToken, CharsToRemove, "",
830 Prefix[LineIndex], InPPDirective, /*Newlines=*/1,
831 /*Spaces=*/IndentAtLineBreak - Prefix[LineIndex].size());
834 BreakableComment::Split BreakableLineCommentSection::getReflowSplit(
835 unsigned LineIndex, llvm::Regex &CommentPragmasRegex) const {
836 if (!mayReflow(LineIndex, CommentPragmasRegex))
837 return Split(StringRef::npos, 0);
839 size_t Trimmed = Content[LineIndex].find_first_not_of(Blanks);
841 // In a line comment section each line is a separate token; thus, after a
842 // split we replace all whitespace before the current line comment token
843 // (which does not need to be included in the split), plus the start of the
844 // line up to where the content starts.
845 return Split(0, Trimmed != StringRef::npos ? Trimmed : 0);
848 void BreakableLineCommentSection::reflow(unsigned LineIndex,
849 WhitespaceManager &Whitespaces) const {
850 if (LineIndex > 0 && Tokens[LineIndex] != Tokens[LineIndex - 1]) {
851 // Reflow happens between tokens. Replace the whitespace between the
852 // tokens by the empty string.
853 Whitespaces.replaceWhitespace(
854 *Tokens[LineIndex], /*Newlines=*/0, /*Spaces=*/0,
855 /*StartOfTokenColumn=*/StartColumn, /*InPPDirective=*/false);
856 } else if (LineIndex > 0) {
857 // In case we're reflowing after the '\' in:
862 // the reflow happens inside the single comment token (it is a single line
863 // comment with an unescaped newline).
864 // Replace the whitespace between the '\' and '//' with the empty string.
866 // Offset points to after the '\' relative to start of the token.
867 unsigned Offset = Lines[LineIndex - 1].data() +
868 Lines[LineIndex - 1].size() -
869 tokenAt(LineIndex - 1).TokenText.data();
870 // WhitespaceLength is the number of chars between the '\' and the '//' on
872 unsigned WhitespaceLength =
873 Lines[LineIndex].data() - tokenAt(LineIndex).TokenText.data() - Offset;
874 Whitespaces.replaceWhitespaceInToken(*Tokens[LineIndex],
876 /*ReplaceChars=*/WhitespaceLength,
877 /*PreviousPostfix=*/"",
878 /*CurrentPrefix=*/"",
879 /*InPPDirective=*/false,
884 // Replace the indent and prefix of the token with the reflow prefix.
886 Lines[LineIndex].data() - tokenAt(LineIndex).TokenText.data();
887 unsigned WhitespaceLength =
888 Content[LineIndex].data() - Lines[LineIndex].data();
889 Whitespaces.replaceWhitespaceInToken(*Tokens[LineIndex],
891 /*ReplaceChars=*/WhitespaceLength,
892 /*PreviousPostfix=*/"",
893 /*CurrentPrefix=*/ReflowPrefix,
894 /*InPPDirective=*/false,
899 void BreakableLineCommentSection::adaptStartOfLine(
900 unsigned LineIndex, WhitespaceManager &Whitespaces) const {
901 // If this is the first line of a token, we need to inform Whitespace Manager
902 // about it: either adapt the whitespace range preceding it, or mark it as an
903 // untouchable token.
904 // This happens for instance here:
907 if (LineIndex > 0 && Tokens[LineIndex] != Tokens[LineIndex - 1]) {
908 // This is the first line for the current token, but no reflow with the
909 // previous token is necessary. However, we still may need to adjust the
910 // start column. Note that ContentColumn[LineIndex] is the expected
911 // content column after a possible update to the prefix, hence the prefix
912 // length change is included.
913 unsigned LineColumn =
914 ContentColumn[LineIndex] -
915 (Content[LineIndex].data() - Lines[LineIndex].data()) +
916 (OriginalPrefix[LineIndex].size() - Prefix[LineIndex].size());
918 // We always want to create a replacement instead of adding an untouchable
919 // token, even if LineColumn is the same as the original column of the
920 // token. This is because WhitespaceManager doesn't align trailing
921 // comments if they are untouchable.
922 Whitespaces.replaceWhitespace(*Tokens[LineIndex],
924 /*Spaces=*/LineColumn,
925 /*StartOfTokenColumn=*/LineColumn,
926 /*InPPDirective=*/false);
928 if (OriginalPrefix[LineIndex] != Prefix[LineIndex]) {
929 // Adjust the prefix if necessary.
931 // Take care of the space possibly introduced after a decoration.
932 assert(Prefix[LineIndex] == (OriginalPrefix[LineIndex] + " ").str() &&
933 "Expecting a line comment prefix to differ from original by at most "
935 Whitespaces.replaceWhitespaceInToken(
936 tokenAt(LineIndex), OriginalPrefix[LineIndex].size(), 0, "", "",
937 /*InPPDirective=*/false, /*Newlines=*/0, /*Spaces=*/1);
941 void BreakableLineCommentSection::updateNextToken(LineState &State) const {
943 State.NextToken = LastLineTok->Next;
947 bool BreakableLineCommentSection::mayReflow(
948 unsigned LineIndex, llvm::Regex &CommentPragmasRegex) const {
949 // Line comments have the indent as part of the prefix, so we need to
950 // recompute the start of the line.
951 StringRef IndentContent = Content[LineIndex];
952 if (Lines[LineIndex].startswith("//")) {
953 IndentContent = Lines[LineIndex].substr(2);
955 // FIXME: Decide whether we want to reflow non-regular indents:
956 // Currently, we only reflow when the OriginalPrefix[LineIndex] matches the
957 // OriginalPrefix[LineIndex-1]. That means we don't reflow
958 // // text that protrudes
959 // // into text with different indent
960 // We do reflow in that case in block comments.
961 return LineIndex > 0 && !CommentPragmasRegex.match(IndentContent) &&
962 mayReflowContent(Content[LineIndex]) && !Tok.Finalized &&
963 !switchesFormatting(tokenAt(LineIndex)) &&
964 OriginalPrefix[LineIndex] == OriginalPrefix[LineIndex - 1];
967 } // namespace format