1 //===--- BreakableToken.cpp - Format C++ code -----------------------------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 ///
10 /// \file
11 /// \brief Contains implementation of BreakableToken class and classes derived
12 /// from it.
13 ///
14 //===----------------------------------------------------------------------===//
15 
16 #define DEBUG_TYPE "format-token-breaker"
17 
18 #include "BreakableToken.h"
19 #include "clang/Basic/CharInfo.h"
20 #include "clang/Format/Format.h"
21 #include "llvm/ADT/STLExtras.h"
22 #include "llvm/Support/Debug.h"
23 #include <algorithm>
24 
25 namespace clang {
26 namespace format {
27 
28 static const char *const Blanks = " \t\v\f\r";
29 static bool IsBlank(char C) {
30   switch (C) {
31   case ' ':
32   case '\t':
33   case '\v':
34   case '\f':
35   case '\r':
36     return true;
37   default:
38     return false;
39   }
40 }
41 
42 static BreakableToken::Split getCommentSplit(StringRef Text,
43                                              unsigned ContentStartColumn,
44                                              unsigned ColumnLimit,
45                                              unsigned TabWidth,
46                                              encoding::Encoding Encoding) {
47   if (ColumnLimit <= ContentStartColumn + 1)
48     return BreakableToken::Split(StringRef::npos, 0);
49 
50   unsigned MaxSplit = ColumnLimit - ContentStartColumn + 1;
51   unsigned MaxSplitBytes = 0;
52 
53   for (unsigned NumChars = 0;
54        NumChars < MaxSplit && MaxSplitBytes < Text.size();) {
55     unsigned BytesInChar =
56         encoding::getCodePointNumBytes(Text[MaxSplitBytes], Encoding);
57     NumChars +=
58         encoding::columnWidthWithTabs(Text.substr(MaxSplitBytes, BytesInChar),
59                                       ContentStartColumn, TabWidth, Encoding);
60     MaxSplitBytes += BytesInChar;
61   }
62 
63   StringRef::size_type SpaceOffset = Text.find_last_of(Blanks, MaxSplitBytes);
64   if (SpaceOffset == StringRef::npos ||
65       // Don't break at leading whitespace.
66       Text.find_last_not_of(Blanks, SpaceOffset) == StringRef::npos) {
67     // Make sure that we don't break at leading whitespace that
68     // reaches past MaxSplit.
69     StringRef::size_type FirstNonWhitespace = Text.find_first_not_of(Blanks);
70     if (FirstNonWhitespace == StringRef::npos)
71       // If the comment is only whitespace, we cannot split.
72       return BreakableToken::Split(StringRef::npos, 0);
73     SpaceOffset = Text.find_first_of(
74         Blanks, std::max<unsigned>(MaxSplitBytes, FirstNonWhitespace));
75   }
76   if (SpaceOffset != StringRef::npos && SpaceOffset != 0) {
77     StringRef BeforeCut = Text.substr(0, SpaceOffset).rtrim(Blanks);
78     StringRef AfterCut = Text.substr(SpaceOffset).ltrim(Blanks);
79     return BreakableToken::Split(BeforeCut.size(),
80                                  AfterCut.begin() - BeforeCut.end());
81   }
82   return BreakableToken::Split(StringRef::npos, 0);
83 }
84 
85 static BreakableToken::Split getStringSplit(StringRef Text,
86                                             unsigned UsedColumns,
87                                             unsigned ColumnLimit,
88                                             unsigned TabWidth,
89                                             encoding::Encoding Encoding) {
90   // FIXME: Reduce unit test case.
91   if (Text.empty())
92     return BreakableToken::Split(StringRef::npos, 0);
93   if (ColumnLimit <= UsedColumns)
94     return BreakableToken::Split(StringRef::npos, 0);
95   unsigned MaxSplit = ColumnLimit - UsedColumns;
96   StringRef::size_type SpaceOffset = 0;
97   StringRef::size_type SlashOffset = 0;
98   StringRef::size_type WordStartOffset = 0;
99   StringRef::size_type SplitPoint = 0;
100   for (unsigned Chars = 0;;) {
101     unsigned Advance;
102     if (Text[0] == '\\') {
103       Advance = encoding::getEscapeSequenceLength(Text);
104       Chars += Advance;
105     } else {
106       Advance = encoding::getCodePointNumBytes(Text[0], Encoding);
107       Chars += encoding::columnWidthWithTabs(
108           Text.substr(0, Advance), UsedColumns + Chars, TabWidth, Encoding);
109     }
110 
111     if (Chars > MaxSplit || Text.size() == Advance)
112       break;
113 
114     if (IsBlank(Text[0]))
115       SpaceOffset = SplitPoint;
116     if (Text[0] == '/')
117       SlashOffset = SplitPoint;
118     if (Advance == 1 && !isAlphanumeric(Text[0]))
119       WordStartOffset = SplitPoint;
120 
121     SplitPoint += Advance;
122     Text = Text.substr(Advance);
123   }
124 
125   if (SpaceOffset != 0)
126     return BreakableToken::Split(SpaceOffset + 1, 0);
127   if (SlashOffset != 0)
128     return BreakableToken::Split(SlashOffset + 1, 0);
129   if (WordStartOffset != 0)
130     return BreakableToken::Split(WordStartOffset + 1, 0);
131   if (SplitPoint != 0)
132     return BreakableToken::Split(SplitPoint, 0);
133   return BreakableToken::Split(StringRef::npos, 0);
134 }
135 
136 unsigned BreakableSingleLineToken::getLineCount() const { return 1; }
137 
138 unsigned BreakableSingleLineToken::getLineLengthAfterSplit(
139     unsigned LineIndex, unsigned Offset, StringRef::size_type Length) const {
140   return StartColumn + Prefix.size() + Postfix.size() +
141          encoding::columnWidthWithTabs(Line.substr(Offset, Length),
142                                        StartColumn + Prefix.size(),
143                                        Style.TabWidth, Encoding);
144 }
145 
146 BreakableSingleLineToken::BreakableSingleLineToken(
147     const FormatToken &Tok, unsigned IndentLevel, unsigned StartColumn,
148     StringRef Prefix, StringRef Postfix, bool InPPDirective,
149     encoding::Encoding Encoding, const FormatStyle &Style)
150     : BreakableToken(Tok, IndentLevel, InPPDirective, Encoding, Style),
151       StartColumn(StartColumn), Prefix(Prefix), Postfix(Postfix) {
152   assert(Tok.TokenText.endswith(Postfix));
153   Line = Tok.TokenText.substr(
154       Prefix.size(), Tok.TokenText.size() - Prefix.size() - Postfix.size());
155 }
156 
157 BreakableStringLiteral::BreakableStringLiteral(
158     const FormatToken &Tok, unsigned IndentLevel, unsigned StartColumn,
159     StringRef Prefix, StringRef Postfix, bool InPPDirective,
160     encoding::Encoding Encoding, const FormatStyle &Style)
161     : BreakableSingleLineToken(Tok, IndentLevel, StartColumn, Prefix, Postfix,
162                                InPPDirective, Encoding, Style) {}
163 
164 BreakableToken::Split
165 BreakableStringLiteral::getSplit(unsigned LineIndex, unsigned TailOffset,
166                                  unsigned ColumnLimit) const {
167   return getStringSplit(Line.substr(TailOffset),
168                         StartColumn + Prefix.size() + Postfix.size(),
169                         ColumnLimit, Style.TabWidth, Encoding);
170 }
171 
172 void BreakableStringLiteral::insertBreak(unsigned LineIndex,
173                                          unsigned TailOffset, Split Split,
174                                          WhitespaceManager &Whitespaces) {
175   unsigned LeadingSpaces = StartColumn;
176   // The '@' of an ObjC string literal (@"Test") does not become part of the
177   // string token.
178   // FIXME: It might be a cleaner solution to merge the tokens as a
179   // precomputation step.
180   if (Prefix.startswith("@"))
181     --LeadingSpaces;
182   Whitespaces.replaceWhitespaceInToken(
183       Tok, Prefix.size() + TailOffset + Split.first, Split.second, Postfix,
184       Prefix, InPPDirective, 1, IndentLevel, LeadingSpaces);
185 }
186 
187 static StringRef getLineCommentIndentPrefix(StringRef Comment) {
188   static const char *const KnownPrefixes[] = { "///", "//" };
189   StringRef LongestPrefix;
190   for (StringRef KnownPrefix : KnownPrefixes) {
191     if (Comment.startswith(KnownPrefix)) {
192       size_t PrefixLength = KnownPrefix.size();
193       while (PrefixLength < Comment.size() && Comment[PrefixLength] == ' ')
194         ++PrefixLength;
195       if (PrefixLength > LongestPrefix.size())
196         LongestPrefix = Comment.substr(0, PrefixLength);
197     }
198   }
199   return LongestPrefix;
200 }
201 
202 BreakableLineComment::BreakableLineComment(
203     const FormatToken &Token, unsigned IndentLevel, unsigned StartColumn,
204     bool InPPDirective, encoding::Encoding Encoding, const FormatStyle &Style)
205     : BreakableSingleLineToken(Token, IndentLevel, StartColumn,
206                                getLineCommentIndentPrefix(Token.TokenText), "",
207                                InPPDirective, Encoding, Style) {
208   OriginalPrefix = Prefix;
209   if (Token.TokenText.size() > Prefix.size() &&
210       isAlphanumeric(Token.TokenText[Prefix.size()])) {
211     if (Prefix == "//")
212       Prefix = "// ";
213     else if (Prefix == "///")
214       Prefix = "/// ";
215   }
216 }
217 
218 BreakableToken::Split
219 BreakableLineComment::getSplit(unsigned LineIndex, unsigned TailOffset,
220                                unsigned ColumnLimit) const {
221   return getCommentSplit(Line.substr(TailOffset), StartColumn + Prefix.size(),
222                          ColumnLimit, Style.TabWidth, Encoding);
223 }
224 
225 void BreakableLineComment::insertBreak(unsigned LineIndex, unsigned TailOffset,
226                                        Split Split,
227                                        WhitespaceManager &Whitespaces) {
228   Whitespaces.replaceWhitespaceInToken(
229       Tok, OriginalPrefix.size() + TailOffset + Split.first, Split.second,
230       Postfix, Prefix, InPPDirective, /*Newlines=*/1, IndentLevel, StartColumn);
231 }
232 
233 void BreakableLineComment::replaceWhitespace(unsigned LineIndex,
234                                              unsigned TailOffset, Split Split,
235                                              WhitespaceManager &Whitespaces) {
236   Whitespaces.replaceWhitespaceInToken(
237       Tok, OriginalPrefix.size() + TailOffset + Split.first, Split.second, "",
238       "", /*InPPDirective=*/false, /*Newlines=*/0, /*IndentLevel=*/0,
239       /*Spaces=*/1);
240 }
241 
242 void
243 BreakableLineComment::replaceWhitespaceBefore(unsigned LineIndex,
244                                               WhitespaceManager &Whitespaces) {
245   if (OriginalPrefix != Prefix) {
246     Whitespaces.replaceWhitespaceInToken(Tok, OriginalPrefix.size(), 0, "", "",
247                                          /*InPPDirective=*/false,
248                                          /*Newlines=*/0, /*IndentLevel=*/0,
249                                          /*Spaces=*/1);
250   }
251 }
252 
253 BreakableBlockComment::BreakableBlockComment(
254     const FormatToken &Token, unsigned IndentLevel, unsigned StartColumn,
255     unsigned OriginalStartColumn, bool FirstInLine, bool InPPDirective,
256     encoding::Encoding Encoding, const FormatStyle &Style)
257     : BreakableToken(Token, IndentLevel, InPPDirective, Encoding, Style) {
258   StringRef TokenText(Token.TokenText);
259   assert(TokenText.startswith("/*") && TokenText.endswith("*/"));
260   TokenText.substr(2, TokenText.size() - 4).split(Lines, "\n");
261 
262   int IndentDelta = StartColumn - OriginalStartColumn;
263   LeadingWhitespace.resize(Lines.size());
264   StartOfLineColumn.resize(Lines.size());
265   StartOfLineColumn[0] = StartColumn + 2;
266   for (size_t i = 1; i < Lines.size(); ++i)
267     adjustWhitespace(i, IndentDelta);
268 
269   Decoration = "* ";
270   if (Lines.size() == 1 && !FirstInLine) {
271     // Comments for which FirstInLine is false can start on arbitrary column,
272     // and available horizontal space can be too small to align consecutive
273     // lines with the first one.
274     // FIXME: We could, probably, align them to current indentation level, but
275     // now we just wrap them without stars.
276     Decoration = "";
277   }
278   for (size_t i = 1, e = Lines.size(); i < e && !Decoration.empty(); ++i) {
279     // If the last line is empty, the closing "*/" will have a star.
280     if (i + 1 == e && Lines[i].empty())
281       break;
282     while (!Lines[i].startswith(Decoration))
283       Decoration = Decoration.substr(0, Decoration.size() - 1);
284   }
285 
286   LastLineNeedsDecoration = true;
287   IndentAtLineBreak = StartOfLineColumn[0] + 1;
288   for (size_t i = 1; i < Lines.size(); ++i) {
289     if (Lines[i].empty()) {
290       if (i + 1 == Lines.size()) {
291         // Empty last line means that we already have a star as a part of the
292         // trailing */. We also need to preserve whitespace, so that */ is
293         // correctly indented.
294         LastLineNeedsDecoration = false;
295       } else if (Decoration.empty()) {
296         // For all other lines, set the start column to 0 if they're empty, so
297         // we do not insert trailing whitespace anywhere.
298         StartOfLineColumn[i] = 0;
299       }
300       continue;
301     }
302     // The first line already excludes the star.
303     // For all other lines, adjust the line to exclude the star and
304     // (optionally) the first whitespace.
305     StartOfLineColumn[i] += Decoration.size();
306     Lines[i] = Lines[i].substr(Decoration.size());
307     LeadingWhitespace[i] += Decoration.size();
308     IndentAtLineBreak = std::min<int>(IndentAtLineBreak, StartOfLineColumn[i]);
309   }
310   IndentAtLineBreak = std::max<unsigned>(IndentAtLineBreak, Decoration.size());
311   DEBUG({
312     llvm::dbgs() << "IndentAtLineBreak " << IndentAtLineBreak << "\n";
313     for (size_t i = 0; i < Lines.size(); ++i) {
314       llvm::dbgs() << i << " |" << Lines[i] << "| " << LeadingWhitespace[i]
315                    << "\n";
316     }
317   });
318 }
319 
320 void BreakableBlockComment::adjustWhitespace(unsigned LineIndex,
321                                              int IndentDelta) {
322   // When in a preprocessor directive, the trailing backslash in a block comment
323   // is not needed, but can serve a purpose of uniformity with necessary escaped
324   // newlines outside the comment. In this case we remove it here before
325   // trimming the trailing whitespace. The backslash will be re-added later when
326   // inserting a line break.
327   size_t EndOfPreviousLine = Lines[LineIndex - 1].size();
328   if (InPPDirective && Lines[LineIndex - 1].endswith("\\"))
329     --EndOfPreviousLine;
330 
331   // Calculate the end of the non-whitespace text in the previous line.
332   EndOfPreviousLine =
333       Lines[LineIndex - 1].find_last_not_of(Blanks, EndOfPreviousLine);
334   if (EndOfPreviousLine == StringRef::npos)
335     EndOfPreviousLine = 0;
336   else
337     ++EndOfPreviousLine;
338   // Calculate the start of the non-whitespace text in the current line.
339   size_t StartOfLine = Lines[LineIndex].find_first_not_of(Blanks);
340   if (StartOfLine == StringRef::npos)
341     StartOfLine = Lines[LineIndex].size();
342 
343   StringRef Whitespace = Lines[LineIndex].substr(0, StartOfLine);
344   // Adjust Lines to only contain relevant text.
345   Lines[LineIndex - 1] = Lines[LineIndex - 1].substr(0, EndOfPreviousLine);
346   Lines[LineIndex] = Lines[LineIndex].substr(StartOfLine);
347   // Adjust LeadingWhitespace to account all whitespace between the lines
348   // to the current line.
349   LeadingWhitespace[LineIndex] =
350       Lines[LineIndex].begin() - Lines[LineIndex - 1].end();
351 
352   // Adjust the start column uniformly across all lines.
353   StartOfLineColumn[LineIndex] =
354       encoding::columnWidthWithTabs(Whitespace, 0, Style.TabWidth, Encoding) +
355       IndentDelta;
356 }
357 
358 unsigned BreakableBlockComment::getLineCount() const { return Lines.size(); }
359 
360 unsigned BreakableBlockComment::getLineLengthAfterSplit(
361     unsigned LineIndex, unsigned Offset, StringRef::size_type Length) const {
362   unsigned ContentStartColumn = getContentStartColumn(LineIndex, Offset);
363   return ContentStartColumn +
364          encoding::columnWidthWithTabs(Lines[LineIndex].substr(Offset, Length),
365                                        ContentStartColumn, Style.TabWidth,
366                                        Encoding) +
367          // The last line gets a "*/" postfix.
368          (LineIndex + 1 == Lines.size() ? 2 : 0);
369 }
370 
371 BreakableToken::Split
372 BreakableBlockComment::getSplit(unsigned LineIndex, unsigned TailOffset,
373                                 unsigned ColumnLimit) const {
374   return getCommentSplit(Lines[LineIndex].substr(TailOffset),
375                          getContentStartColumn(LineIndex, TailOffset),
376                          ColumnLimit, Style.TabWidth, Encoding);
377 }
378 
379 void BreakableBlockComment::insertBreak(unsigned LineIndex, unsigned TailOffset,
380                                         Split Split,
381                                         WhitespaceManager &Whitespaces) {
382   StringRef Text = Lines[LineIndex].substr(TailOffset);
383   StringRef Prefix = Decoration;
384   if (LineIndex + 1 == Lines.size() &&
385       Text.size() == Split.first + Split.second) {
386     // For the last line we need to break before "*/", but not to add "* ".
387     Prefix = "";
388   }
389 
390   unsigned BreakOffsetInToken =
391       Text.data() - Tok.TokenText.data() + Split.first;
392   unsigned CharsToRemove = Split.second;
393   assert(IndentAtLineBreak >= Decoration.size());
394   Whitespaces.replaceWhitespaceInToken(
395       Tok, BreakOffsetInToken, CharsToRemove, "", Prefix, InPPDirective, 1,
396       IndentLevel, IndentAtLineBreak - Decoration.size());
397 }
398 
399 void BreakableBlockComment::replaceWhitespace(unsigned LineIndex,
400                                               unsigned TailOffset, Split Split,
401                                               WhitespaceManager &Whitespaces) {
402   StringRef Text = Lines[LineIndex].substr(TailOffset);
403   unsigned BreakOffsetInToken =
404       Text.data() - Tok.TokenText.data() + Split.first;
405   unsigned CharsToRemove = Split.second;
406   Whitespaces.replaceWhitespaceInToken(
407       Tok, BreakOffsetInToken, CharsToRemove, "", "", /*InPPDirective=*/false,
408       /*Newlines=*/0, /*IndentLevel=*/0, /*Spaces=*/1);
409 }
410 
411 void
412 BreakableBlockComment::replaceWhitespaceBefore(unsigned LineIndex,
413                                                WhitespaceManager &Whitespaces) {
414   if (LineIndex == 0)
415     return;
416   StringRef Prefix = Decoration;
417   if (Lines[LineIndex].empty()) {
418     if (LineIndex + 1 == Lines.size()) {
419       if (!LastLineNeedsDecoration) {
420         // If the last line was empty, we don't need a prefix, as the */ will
421         // line up with the decoration (if it exists).
422         Prefix = "";
423       }
424     } else if (!Decoration.empty()) {
425       // For other empty lines, if we do have a decoration, adapt it to not
426       // contain a trailing whitespace.
427       Prefix = Prefix.substr(0, 1);
428     }
429   } else {
430     if (StartOfLineColumn[LineIndex] == 1) {
431       // This line starts immediately after the decorating *.
432       Prefix = Prefix.substr(0, 1);
433     }
434   }
435 
436   unsigned WhitespaceOffsetInToken = Lines[LineIndex].data() -
437                                      Tok.TokenText.data() -
438                                      LeadingWhitespace[LineIndex];
439   Whitespaces.replaceWhitespaceInToken(
440       Tok, WhitespaceOffsetInToken, LeadingWhitespace[LineIndex], "", Prefix,
441       InPPDirective, 1, IndentLevel,
442       StartOfLineColumn[LineIndex] - Prefix.size());
443 }
444 
445 unsigned
446 BreakableBlockComment::getContentStartColumn(unsigned LineIndex,
447                                              unsigned TailOffset) const {
448   // If we break, we always break at the predefined indent.
449   if (TailOffset != 0)
450     return IndentAtLineBreak;
451   return std::max(0, StartOfLineColumn[LineIndex]);
452 }
453 
454 } // namespace format
455 } // namespace clang
456