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[] = {
"//",
"#",
"##",
"###",
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);
67 unsigned ContentStartColumn,
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 =
86 ContentStartColumn, TabWidth,
Encoding);
87 MaxSplitBytes += BytesInChar;
90 StringRef::size_type SpaceOffset = Text.find_last_of(Blanks, MaxSplitBytes);
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 ||
101 Text.find_last_not_of(Blanks, SpaceOffset) == StringRef::npos) {
104 StringRef::size_type FirstNonWhitespace = Text.find_first_not_of(Blanks);
105 if (FirstNonWhitespace == StringRef::npos)
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
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] ==
'\\') {
141 Text.substr(0, Advance), UsedColumns + Chars, TabWidth,
Encoding);
144 if (Chars > MaxSplit || Text.size() <= Advance)
148 SpaceOffset = SplitPoint;
150 SlashOffset = SplitPoint;
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);
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");
191 return RemainingTokenColumns + 1 - Split.second;
198 StringRef::size_type Length,
200 llvm_unreachable(
"Getting the length of a part of the string literal " 201 "indicates that the code tries to reflow it.");
222 StartColumn(StartColumn), Prefix(Prefix), Postfix(Postfix),
223 UnbreakableTailLength(UnbreakableTailLength) {
226 Prefix.size(), Tok.
TokenText.size() - Prefix.size() - Postfix.size());
230 unsigned LineIndex,
unsigned TailOffset,
unsigned ColumnLimit,
231 unsigned ContentStartColumn, llvm::Regex &CommentPragmasRegex)
const {
237 unsigned TailOffset, Split Split,
238 unsigned ContentIndent,
250 StartColumn(StartColumn) {}
254 BreakableToken::Split
256 unsigned ColumnLimit,
unsigned ContentStartColumn,
257 llvm::Regex &CommentPragmasRegex)
const {
259 if (CommentPragmasRegex.match(
Content[LineIndex]))
260 return Split(StringRef::npos, 0);
267 unsigned LineIndex,
unsigned TailOffset, Split Split,
269 StringRef
Text =
Content[LineIndex].substr(TailOffset);
274 unsigned BreakOffsetInToken =
276 unsigned CharsToRemove = Split.second;
278 tokenAt(LineIndex), BreakOffsetInToken, CharsToRemove,
"",
"",
287 Content = Content.trim(Blanks);
290 bool hasSpecialMeaningPrefix =
false;
291 for (StringRef Prefix :
292 {
"@",
"TODO",
"FIXME",
"XXX",
"-# ",
"- ",
"+ ",
"* "}) {
293 if (Content.startswith(Prefix)) {
294 hasSpecialMeaningPrefix =
true;
302 static auto *
const kNumberedListRegexp =
new llvm::Regex(
"^[1-9][0-9]?\\. ");
303 hasSpecialMeaningPrefix =
304 hasSpecialMeaningPrefix || kNumberedListRegexp->match(Content);
309 return Content.size() >= 2 && !hasSpecialMeaningPrefix &&
310 !Content.endswith(
"\\") &&
318 unsigned OriginalStartColumn,
bool FirstInLine,
bool InPPDirective,
321 DelimitersOnNewline(
false),
322 UnbreakableTailLength(Token.UnbreakableTailLength) {
323 assert(
Tok.
is(TT_BlockComment) &&
324 "block comment section must start with a block comment");
327 assert(TokenText.startswith(
"/*") && TokenText.endswith(
"*/"));
328 TokenText.substr(2, TokenText.size() - 4).split(
Lines,
"\n");
330 int IndentDelta = StartColumn - OriginalStartColumn;
337 for (
size_t i = 1; i <
Lines.size(); ++i)
338 adjustWhitespace(i, IndentDelta);
342 DecorationColumn = StartColumn + 1;
355 if (
Lines.size() == 1 && !FirstInLine) {
363 for (
size_t i = 1, e =
Lines.size(); i < e && !Decoration.empty(); ++i) {
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;
375 for (
size_t i = 1, e =
Lines.size(); i < e; ++i) {
381 LastLineNeedsDecoration =
false;
383 if (e >= 2 && !Decoration.empty()) {
386 }
else if (Decoration.empty()) {
398 unsigned DecorationSize = Decoration.startswith(
Content[i])
401 if (DecorationSize) {
405 if (!Decoration.startswith(
Content[i]))
409 IndentAtLineBreak = std::max<unsigned>(IndentAtLineBreak, Decoration.size());
414 if ((
Lines[0] ==
"*" ||
Lines[0].startswith(
"* ")) &&
Lines.size() > 1) {
416 DelimitersOnNewline =
true;
417 }
else if (
Lines[0].startswith(
"* ") &&
Lines.size() == 1) {
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] <<
"| " 436 <<
"IN=" << (
Content[i].data() -
Lines[i].data()) <<
"\n";
441 void BreakableBlockComment::adjustWhitespace(
unsigned LineIndex,
448 size_t EndOfPreviousLine =
Lines[LineIndex - 1].size();
454 Lines[LineIndex - 1].find_last_not_of(Blanks, EndOfPreviousLine);
455 if (EndOfPreviousLine == StringRef::npos)
456 EndOfPreviousLine = 0;
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);
466 size_t PreviousContentOffset =
467 Content[LineIndex - 1].data() -
Lines[LineIndex - 1].data();
469 PreviousContentOffset, EndOfPreviousLine - PreviousContentOffset);
470 Content[LineIndex] =
Lines[LineIndex].substr(StartOfLine);
480 StringRef::size_type Length,
482 unsigned LineLength =
488 if (LineIndex + 1 ==
Lines.size()) {
493 if (
Content[LineIndex].substr(Offset, StringRef::npos).empty()) {
494 LineLength -= Decoration.size();
503 return UnbreakableTailLength +
510 return IndentAtLineBreak;
514 const llvm::StringSet<>
516 "@param",
"@return",
"@returns",
"@throws",
"@type",
"@template",
517 "@see",
"@deprecated",
"@define",
"@exports",
"@mods",
"@private",
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));
540 Split Split,
unsigned ContentIndent,
542 StringRef
Text =
Content[LineIndex].substr(TailOffset);
543 StringRef Prefix = Decoration;
547 unsigned LocalIndentAtLineBreak = IndentAtLineBreak;
548 if (LineIndex + 1 ==
Lines.size() &&
549 Text.size() == Split.first + Split.second) {
552 if (LocalIndentAtLineBreak >= 2)
553 LocalIndentAtLineBreak -= 2;
557 unsigned BreakOffsetInToken =
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 +=
" ";
565 tokenAt(LineIndex), BreakOffsetInToken, CharsToRemove,
"",
567 LocalIndentAtLineBreak + ContentIndent -
568 PrefixWithTrailingIndent.size());
571 BreakableToken::Split
573 llvm::Regex &CommentPragmasRegex)
const {
574 if (!
mayReflow(LineIndex, CommentPragmasRegex))
575 return Split(StringRef::npos, 0);
579 size_t Trimmed =
Content[LineIndex].find_first_not_of(Blanks);
582 if (PreviousContentIndent && Trimmed != StringRef::npos &&
583 Trimmed != PreviousContentIndent)
584 return Split(StringRef::npos, 0);
587 return Split(0, Trimmed != StringRef::npos ? Trimmed : 0);
592 return DelimitersOnNewline &&
593 Lines[0].substr(1).find_first_not_of(Blanks) != StringRef::npos;
598 StringRef TrimmedContent =
Content[LineIndex].ltrim(Blanks);
601 "Reflowing whitespace within a token");
604 unsigned WhitespaceOffsetInToken =
Content[LineIndex - 1].data() +
605 Content[LineIndex - 1].size() -
607 unsigned WhitespaceLength = TrimmedContent.data() -
609 WhitespaceOffsetInToken;
611 tokenAt(LineIndex), WhitespaceOffsetInToken,
612 WhitespaceLength,
"",
619 if (LineIndex == 0) {
620 if (DelimitersOnNewline) {
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), 0,
632 StringRef Prefix = Decoration;
633 if (
Content[LineIndex].empty()) {
634 if (LineIndex + 1 ==
Lines.size()) {
635 if (!LastLineNeedsDecoration) {
640 }
else if (!Decoration.empty()) {
643 Prefix = Prefix.substr(0, 1);
648 Prefix = Prefix.substr(0, 1);
653 unsigned WhitespaceOffsetInToken =
Content[LineIndex - 1].data() +
654 Content[LineIndex - 1].size() -
656 unsigned WhitespaceLength =
Content[LineIndex].data() -
658 WhitespaceOffsetInToken;
660 tokenAt(LineIndex), WhitespaceOffsetInToken, WhitespaceLength,
"", Prefix,
664 BreakableToken::Split
666 if (DelimitersOnNewline) {
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);
679 llvm::Regex &CommentPragmasRegex)
const {
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) &&
693 unsigned OriginalStartColumn,
bool FirstInLine,
bool InPPDirective,
696 assert(
Tok.
is(TT_LineComment) &&
697 "line comment section must start with a line comment");
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");
710 OriginalContentColumn.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) {
718 StringRef IndentPrefix =
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() &&
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] ==
"#" &&
742 OriginalContentColumn[i] =
751 size_t EndOfLine =
Content[i].find_last_not_of(Blanks);
752 if (EndOfLine == StringRef::npos)
758 LineTok = CurrentTok->Next;
759 if (CurrentTok->Next && !CurrentTok->Next->ContinuesLineCommentSection) {
781 StringRef::size_type Length,
791 return OriginalContentColumn[LineIndex];
796 unsigned LineIndex,
unsigned TailOffset, Split Split,
798 StringRef
Text =
Content[LineIndex].substr(TailOffset);
801 unsigned BreakOffsetInToken =
803 unsigned CharsToRemove = Split.second;
806 unsigned IndentAtLineBreak = OriginalContentColumn[LineIndex] +
807 Prefix[LineIndex].size() -
808 OriginalPrefix[LineIndex].size();
809 assert(IndentAtLineBreak >= Prefix[LineIndex].size());
811 tokenAt(LineIndex), BreakOffsetInToken, CharsToRemove,
"",
813 IndentAtLineBreak - Prefix[LineIndex].size());
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);
827 return Split(0, Trimmed != StringRef::npos ? Trimmed : 0);
832 if (LineIndex > 0 &&
Tokens[LineIndex] !=
Tokens[LineIndex - 1]) {
838 }
else if (LineIndex > 0) {
850 Lines[LineIndex - 1].size() -
854 unsigned WhitespaceLength =
869 unsigned WhitespaceLength =
889 if (LineIndex > 0 &&
Tokens[LineIndex] !=
Tokens[LineIndex - 1]) {
895 unsigned LineColumn =
898 (OriginalPrefix[LineIndex].size() - Prefix[LineIndex].size());
910 if (OriginalPrefix[LineIndex] != Prefix[LineIndex]) {
914 assert(Prefix[LineIndex] == (OriginalPrefix[LineIndex] +
" ").str() &&
915 "Expecting a line comment prefix to differ from original by at most " 918 tokenAt(LineIndex), OriginalPrefix[LineIndex].size(), 0,
"",
"",
930 unsigned LineIndex, llvm::Regex &CommentPragmasRegex)
const {
933 StringRef IndentContent =
Content[LineIndex];
934 if (
Lines[LineIndex].startswith(
"//")) {
935 IndentContent =
Lines[LineIndex].substr(2);
943 return LineIndex > 0 && !CommentPragmasRegex.match(IndentContent) &&
946 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.
LLVM_READONLY bool isPunctuation(unsigned char c)
Return true if this character is an ASCII punctuation character.
__DEVICE__ int max(int __a, int __b)
This file implements an indenter that manages the indentation of continuations.