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";
45 static const char *
const KnownCStylePrefixes[] = {
"///<",
"//!<",
"///",
"//",
47 static const char *
const KnownTextProtoPrefixes[] = {
"//",
"#"};
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);
66 unsigned ContentStartColumn,
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 =
84 ContentStartColumn, TabWidth,
Encoding);
85 MaxSplitBytes += BytesInChar;
88 StringRef::size_type SpaceOffset = Text.find_last_of(Blanks, MaxSplitBytes);
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 ||
99 Text.find_last_not_of(Blanks, SpaceOffset) == StringRef::npos) {
102 StringRef::size_type FirstNonWhitespace = Text.find_first_not_of(Blanks);
103 if (FirstNonWhitespace == StringRef::npos)
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
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] ==
'\\') {
139 Text.substr(0, Advance), UsedColumns + Chars, TabWidth,
Encoding);
142 if (Chars > MaxSplit || Text.size() <= Advance)
146 SpaceOffset = SplitPoint;
148 SlashOffset = SplitPoint;
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);
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");
189 return RemainingTokenColumns + 1 - Split.second;
196 StringRef::size_type Length,
198 llvm_unreachable(
"Getting the length of a part of the string literal " 199 "indicates that the code tries to reflow it.");
220 StartColumn(StartColumn), Prefix(Prefix), Postfix(Postfix),
224 Prefix.size(), Tok.
TokenText.size() - Prefix.size() - Postfix.size());
228 unsigned LineIndex,
unsigned TailOffset,
unsigned ColumnLimit,
229 unsigned ContentStartColumn, llvm::Regex &CommentPragmasRegex)
const {
235 unsigned TailOffset, Split Split,
247 StartColumn(StartColumn) {}
251 BreakableToken::Split
253 unsigned ColumnLimit,
unsigned ContentStartColumn,
254 llvm::Regex &CommentPragmasRegex)
const {
256 if (CommentPragmasRegex.match(
Content[LineIndex]))
257 return Split(StringRef::npos, 0);
264 unsigned LineIndex,
unsigned TailOffset, Split Split,
266 StringRef
Text =
Content[LineIndex].substr(TailOffset);
271 unsigned BreakOffsetInToken =
273 unsigned CharsToRemove = Split.second;
275 tokenAt(LineIndex), BreakOffsetInToken, CharsToRemove,
"",
"",
284 Content = Content.trim(Blanks);
288 "@",
"TODO",
"FIXME",
"XXX",
"-# ",
"- ",
"+ ",
"* "};
289 bool hasSpecialMeaningPrefix =
false;
290 for (StringRef Prefix : kSpecialMeaningPrefixes) {
291 if (Content.startswith(Prefix)) {
292 hasSpecialMeaningPrefix =
true;
300 static llvm::Regex kNumberedListRegexp = llvm::Regex(
"^[1-9][0-9]?\\. ");
301 hasSpecialMeaningPrefix =
302 hasSpecialMeaningPrefix || kNumberedListRegexp.match(Content);
307 return Content.size() >= 2 && !hasSpecialMeaningPrefix &&
308 !Content.endswith(
"\\") &&
316 unsigned OriginalStartColumn,
bool FirstInLine,
bool InPPDirective,
319 DelimitersOnNewline(
false),
320 UnbreakableTailLength(Token.UnbreakableTailLength) {
321 assert(
Tok.
is(TT_BlockComment) &&
322 "block comment section must start with a block comment");
325 assert(TokenText.startswith(
"/*") && TokenText.endswith(
"*/"));
326 TokenText.substr(2, TokenText.size() - 4).split(
Lines,
"\n");
328 int IndentDelta = StartColumn - OriginalStartColumn;
335 for (
size_t i = 1; i <
Lines.size(); ++i)
336 adjustWhitespace(i, IndentDelta);
340 DecorationColumn = StartColumn + 1;
353 if (
Lines.size() == 1 && !FirstInLine) {
361 for (
size_t i = 1, e =
Lines.size(); i < e && !Decoration.empty(); ++i) {
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;
373 for (
size_t i = 1, e =
Lines.size(); i < e; ++i) {
379 LastLineNeedsDecoration =
false;
381 if (e >= 2 && !Decoration.empty()) {
384 }
else if (Decoration.empty()) {
396 unsigned DecorationSize = Decoration.startswith(
Content[i])
399 if (DecorationSize) {
403 if (!Decoration.startswith(
Content[i]))
407 IndentAtLineBreak = std::max<unsigned>(IndentAtLineBreak, Decoration.size());
412 if ((
Lines[0] ==
"*" ||
Lines[0].startswith(
"* ")) &&
Lines.size() > 1) {
414 DelimitersOnNewline =
true;
415 }
else if (
Lines[0].startswith(
"* ") &&
Lines.size() == 1) {
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] <<
"| " 434 <<
"IN=" << (
Content[i].data() -
Lines[i].data()) <<
"\n";
439 void BreakableBlockComment::adjustWhitespace(
unsigned LineIndex,
446 size_t EndOfPreviousLine =
Lines[LineIndex - 1].size();
452 Lines[LineIndex - 1].find_last_not_of(Blanks, EndOfPreviousLine);
453 if (EndOfPreviousLine == StringRef::npos)
454 EndOfPreviousLine = 0;
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);
464 size_t PreviousContentOffset =
465 Content[LineIndex - 1].data() -
Lines[LineIndex - 1].data();
467 PreviousContentOffset, EndOfPreviousLine - PreviousContentOffset);
468 Content[LineIndex] =
Lines[LineIndex].substr(StartOfLine);
478 StringRef::size_type Length,
480 unsigned LineLength =
486 if (LineIndex + 1 ==
Lines.size()) {
491 if (
Content[LineIndex].substr(Offset, StringRef::npos).empty()) {
492 LineLength -= Decoration.size();
501 return UnbreakableTailLength +
508 return IndentAtLineBreak;
515 StringRef
Text =
Content[LineIndex].substr(TailOffset);
516 StringRef Prefix = Decoration;
520 unsigned LocalIndentAtLineBreak = IndentAtLineBreak;
521 if (LineIndex + 1 ==
Lines.size() &&
522 Text.size() == Split.first + Split.second) {
525 if (LocalIndentAtLineBreak >= 2)
526 LocalIndentAtLineBreak -= 2;
530 unsigned BreakOffsetInToken =
532 unsigned CharsToRemove = Split.second;
533 assert(LocalIndentAtLineBreak >= Prefix.size());
535 tokenAt(LineIndex), BreakOffsetInToken, CharsToRemove,
"", Prefix,
537 LocalIndentAtLineBreak - Prefix.size());
540 BreakableToken::Split
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);
552 return DelimitersOnNewline &&
553 Lines[0].substr(1).find_first_not_of(Blanks) != StringRef::npos;
558 StringRef TrimmedContent =
Content[LineIndex].ltrim(Blanks);
561 "Reflowing whitespace within a token");
564 unsigned WhitespaceOffsetInToken =
Content[LineIndex - 1].data() +
565 Content[LineIndex - 1].size() -
567 unsigned WhitespaceLength = TrimmedContent.data() -
569 WhitespaceOffsetInToken;
571 tokenAt(LineIndex), WhitespaceOffsetInToken,
572 WhitespaceLength,
"",
579 if (LineIndex == 0) {
580 if (DelimitersOnNewline) {
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);
591 StringRef Prefix = Decoration;
592 if (
Content[LineIndex].empty()) {
593 if (LineIndex + 1 ==
Lines.size()) {
594 if (!LastLineNeedsDecoration) {
599 }
else if (!Decoration.empty()) {
602 Prefix = Prefix.substr(0, 1);
607 Prefix = Prefix.substr(0, 1);
612 unsigned WhitespaceOffsetInToken =
Content[LineIndex - 1].data() +
613 Content[LineIndex - 1].size() -
615 unsigned WhitespaceLength =
Content[LineIndex].data() -
617 WhitespaceOffsetInToken;
619 tokenAt(LineIndex), WhitespaceOffsetInToken, WhitespaceLength,
"", Prefix,
623 BreakableToken::Split
625 if (DelimitersOnNewline) {
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);
638 llvm::Regex &CommentPragmasRegex)
const {
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) &&
652 unsigned OriginalStartColumn,
bool FirstInLine,
bool InPPDirective,
655 assert(
Tok.
is(TT_LineComment) &&
656 "line comment section must start with a line comment");
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");
669 OriginalContentColumn.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) {
677 StringRef IndentPrefix =
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() &&
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] ==
"#" &&
701 OriginalContentColumn[i] =
710 size_t EndOfLine =
Content[i].find_last_not_of(Blanks);
711 if (EndOfLine == StringRef::npos)
717 LineTok = CurrentTok->Next;
718 if (CurrentTok->Next && !CurrentTok->Next->ContinuesLineCommentSection) {
740 StringRef::size_type Length,
750 return OriginalContentColumn[LineIndex];
755 unsigned LineIndex,
unsigned TailOffset, Split Split,
757 StringRef
Text =
Content[LineIndex].substr(TailOffset);
760 unsigned BreakOffsetInToken =
762 unsigned CharsToRemove = Split.second;
765 unsigned IndentAtLineBreak = OriginalContentColumn[LineIndex] +
766 Prefix[LineIndex].size() -
767 OriginalPrefix[LineIndex].size();
768 assert(IndentAtLineBreak >= Prefix[LineIndex].size());
770 tokenAt(LineIndex), BreakOffsetInToken, CharsToRemove,
"",
772 IndentAtLineBreak - Prefix[LineIndex].size());
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);
786 return Split(0, Trimmed != StringRef::npos ? Trimmed : 0);
797 unsigned WhitespaceLength =
817 if (LineIndex > 0 &&
Tokens[LineIndex] !=
Tokens[LineIndex - 1]) {
823 unsigned LineColumn =
826 (OriginalPrefix[LineIndex].size() - Prefix[LineIndex].size());
838 if (OriginalPrefix[LineIndex] != Prefix[LineIndex]) {
842 assert(Prefix[LineIndex] == (OriginalPrefix[LineIndex] +
" ").str() &&
843 "Expecting a line comment prefix to differ from original by at most " 846 tokenAt(LineIndex), OriginalPrefix[LineIndex].size(), 0,
"",
"",
858 unsigned LineIndex, llvm::Regex &CommentPragmasRegex)
const {
861 StringRef IndentContent =
Content[LineIndex];
862 if (
Lines[LineIndex].startswith(
"//")) {
863 IndentContent =
Lines[LineIndex].substr(2);
871 return LineIndex > 0 && !CommentPragmasRegex.match(IndentContent) &&
874 OriginalPrefix[LineIndex] == OriginalPrefix[LineIndex - 1];
Declares BreakableToken, BreakableStringLiteral, BreakableComment, BreakableBlockComment and Breakabl...
Token - This structure provides full information about a lexed token.
const AnnotatedLine * Line
LLVM_READONLY bool isAlphanumeric(unsigned char c)
Return true if this character is an ASCII letter or digit: [a-zA-Z0-9].
Dataflow Directional Tag Classes.
char __ovld __cnfn max(char x, char y)
Returns y if x < y, otherwise it returns x.
LLVM_READONLY bool isPunctuation(unsigned char c)
Return true if this character is an ASCII punctuation character.
This file implements an indenter that manages the indentation of continuations.