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 #include "BreakableToken.h"
17 #include "clang/Basic/CharInfo.h"
18 #include "clang/Format/Format.h"
19 #include "llvm/ADT/STLExtras.h"
20 #include "llvm/Support/Debug.h"
21 #include <algorithm>
22 
23 #define DEBUG_TYPE "format-token-breaker"
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
86 getStringSplit(StringRef Text, unsigned UsedColumns, unsigned ColumnLimit,
87                unsigned TabWidth, encoding::Encoding Encoding) {
88   // FIXME: Reduce unit test case.
89   if (Text.empty())
90     return BreakableToken::Split(StringRef::npos, 0);
91   if (ColumnLimit <= UsedColumns)
92     return BreakableToken::Split(StringRef::npos, 0);
93   unsigned MaxSplit = ColumnLimit - UsedColumns;
94   StringRef::size_type SpaceOffset = 0;
95   StringRef::size_type SlashOffset = 0;
96   StringRef::size_type WordStartOffset = 0;
97   StringRef::size_type SplitPoint = 0;
98   for (unsigned Chars = 0;;) {
99     unsigned Advance;
100     if (Text[0] == '\\') {
101       Advance = encoding::getEscapeSequenceLength(Text);
102       Chars += Advance;
103     } else {
104       Advance = encoding::getCodePointNumBytes(Text[0], Encoding);
105       Chars += encoding::columnWidthWithTabs(
106           Text.substr(0, Advance), UsedColumns + Chars, TabWidth, Encoding);
107     }
108 
109     if (Chars > MaxSplit || Text.size() <= Advance)
110       break;
111 
112     if (IsBlank(Text[0]))
113       SpaceOffset = SplitPoint;
114     if (Text[0] == '/')
115       SlashOffset = SplitPoint;
116     if (Advance == 1 && !isAlphanumeric(Text[0]))
117       WordStartOffset = SplitPoint;
118 
119     SplitPoint += Advance;
120     Text = Text.substr(Advance);
121   }
122 
123   if (SpaceOffset != 0)
124     return BreakableToken::Split(SpaceOffset + 1, 0);
125   if (SlashOffset != 0)
126     return BreakableToken::Split(SlashOffset + 1, 0);
127   if (WordStartOffset != 0)
128     return BreakableToken::Split(WordStartOffset + 1, 0);
129   if (SplitPoint != 0)
130     return BreakableToken::Split(SplitPoint, 0);
131   return BreakableToken::Split(StringRef::npos, 0);
132 }
133 
134 unsigned BreakableSingleLineToken::getLineCount() const { return 1; }
135 
136 unsigned BreakableSingleLineToken::getLineLengthAfterSplit(
137     unsigned LineIndex, unsigned Offset, StringRef::size_type Length) const {
138   return StartColumn + Prefix.size() + Postfix.size() +
139          encoding::columnWidthWithTabs(Line.substr(Offset, Length),
140                                        StartColumn + Prefix.size(),
141                                        Style.TabWidth, Encoding);
142 }
143 
144 BreakableSingleLineToken::BreakableSingleLineToken(
145     const FormatToken &Tok, unsigned IndentLevel, unsigned StartColumn,
146     StringRef Prefix, StringRef Postfix, bool InPPDirective,
147     encoding::Encoding Encoding, const FormatStyle &Style)
148     : BreakableToken(Tok, IndentLevel, InPPDirective, Encoding, Style),
149       StartColumn(StartColumn), Prefix(Prefix), Postfix(Postfix) {
150   assert(Tok.TokenText.endswith(Postfix));
151   Line = Tok.TokenText.substr(
152       Prefix.size(), Tok.TokenText.size() - Prefix.size() - Postfix.size());
153 }
154 
155 BreakableStringLiteral::BreakableStringLiteral(
156     const FormatToken &Tok, unsigned IndentLevel, unsigned StartColumn,
157     StringRef Prefix, StringRef Postfix, bool InPPDirective,
158     encoding::Encoding Encoding, const FormatStyle &Style)
159     : BreakableSingleLineToken(Tok, IndentLevel, StartColumn, Prefix, Postfix,
160                                InPPDirective, Encoding, Style) {}
161 
162 BreakableToken::Split
163 BreakableStringLiteral::getSplit(unsigned LineIndex, unsigned TailOffset,
164                                  unsigned ColumnLimit) const {
165   return getStringSplit(Line.substr(TailOffset),
166                         StartColumn + Prefix.size() + Postfix.size(),
167                         ColumnLimit, Style.TabWidth, Encoding);
168 }
169 
170 void BreakableStringLiteral::insertBreak(unsigned LineIndex,
171                                          unsigned TailOffset, Split Split,
172                                          WhitespaceManager &Whitespaces) {
173   unsigned LeadingSpaces = StartColumn;
174   // The '@' of an ObjC string literal (@"Test") does not become part of the
175   // string token.
176   // FIXME: It might be a cleaner solution to merge the tokens as a
177   // precomputation step.
178   if (Prefix.startswith("@"))
179     --LeadingSpaces;
180   Whitespaces.replaceWhitespaceInToken(
181       Tok, Prefix.size() + TailOffset + Split.first, Split.second, Postfix,
182       Prefix, InPPDirective, 1, IndentLevel, LeadingSpaces);
183 }
184 
185 static StringRef getLineCommentIndentPrefix(StringRef Comment) {
186   static const char *const KnownPrefixes[] = { "///", "//", "//!" };
187   StringRef LongestPrefix;
188   for (StringRef KnownPrefix : KnownPrefixes) {
189     if (Comment.startswith(KnownPrefix)) {
190       size_t PrefixLength = KnownPrefix.size();
191       while (PrefixLength < Comment.size() && Comment[PrefixLength] == ' ')
192         ++PrefixLength;
193       if (PrefixLength > LongestPrefix.size())
194         LongestPrefix = Comment.substr(0, PrefixLength);
195     }
196   }
197   return LongestPrefix;
198 }
199 
200 BreakableLineComment::BreakableLineComment(
201     const FormatToken &Token, unsigned IndentLevel, unsigned StartColumn,
202     bool InPPDirective, encoding::Encoding Encoding, const FormatStyle &Style)
203     : BreakableSingleLineToken(Token, IndentLevel, StartColumn,
204                                getLineCommentIndentPrefix(Token.TokenText), "",
205                                InPPDirective, Encoding, Style) {
206   OriginalPrefix = Prefix;
207   if (Token.TokenText.size() > Prefix.size() &&
208       isAlphanumeric(Token.TokenText[Prefix.size()])) {
209     if (Prefix == "//")
210       Prefix = "// ";
211     else 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     if (!Lines[i].empty() && i + 1 != e && Decoration.startswith(Lines[i]))
283       continue;
284     while (!Lines[i].startswith(Decoration))
285       Decoration = Decoration.substr(0, Decoration.size() - 1);
286   }
287 
288   LastLineNeedsDecoration = true;
289   IndentAtLineBreak = StartOfLineColumn[0] + 1;
290   for (size_t i = 1; i < Lines.size(); ++i) {
291     if (Lines[i].empty()) {
292       if (i + 1 == Lines.size()) {
293         // Empty last line means that we already have a star as a part of the
294         // trailing */. We also need to preserve whitespace, so that */ is
295         // correctly indented.
296         LastLineNeedsDecoration = false;
297       } else if (Decoration.empty()) {
298         // For all other lines, set the start column to 0 if they're empty, so
299         // we do not insert trailing whitespace anywhere.
300         StartOfLineColumn[i] = 0;
301       }
302       continue;
303     }
304 
305     // The first line already excludes the star.
306     // For all other lines, adjust the line to exclude the star and
307     // (optionally) the first whitespace.
308     unsigned DecorationSize =
309         Decoration.startswith(Lines[i]) ? Lines[i].size() : Decoration.size();
310     StartOfLineColumn[i] += DecorationSize;
311     Lines[i] = Lines[i].substr(DecorationSize);
312     LeadingWhitespace[i] += DecorationSize;
313     if (!Decoration.startswith(Lines[i]))
314       IndentAtLineBreak =
315           std::min<int>(IndentAtLineBreak, std::max(0, StartOfLineColumn[i]));
316   }
317   IndentAtLineBreak = std::max<unsigned>(IndentAtLineBreak, Decoration.size());
318   DEBUG({
319     llvm::dbgs() << "IndentAtLineBreak " << IndentAtLineBreak << "\n";
320     for (size_t i = 0; i < Lines.size(); ++i) {
321       llvm::dbgs() << i << " |" << Lines[i] << "| " << LeadingWhitespace[i]
322                    << "\n";
323     }
324   });
325 }
326 
327 void BreakableBlockComment::adjustWhitespace(unsigned LineIndex,
328                                              int IndentDelta) {
329   // When in a preprocessor directive, the trailing backslash in a block comment
330   // is not needed, but can serve a purpose of uniformity with necessary escaped
331   // newlines outside the comment. In this case we remove it here before
332   // trimming the trailing whitespace. The backslash will be re-added later when
333   // inserting a line break.
334   size_t EndOfPreviousLine = Lines[LineIndex - 1].size();
335   if (InPPDirective && Lines[LineIndex - 1].endswith("\\"))
336     --EndOfPreviousLine;
337 
338   // Calculate the end of the non-whitespace text in the previous line.
339   EndOfPreviousLine =
340       Lines[LineIndex - 1].find_last_not_of(Blanks, EndOfPreviousLine);
341   if (EndOfPreviousLine == StringRef::npos)
342     EndOfPreviousLine = 0;
343   else
344     ++EndOfPreviousLine;
345   // Calculate the start of the non-whitespace text in the current line.
346   size_t StartOfLine = Lines[LineIndex].find_first_not_of(Blanks);
347   if (StartOfLine == StringRef::npos)
348     StartOfLine = Lines[LineIndex].size();
349 
350   StringRef Whitespace = Lines[LineIndex].substr(0, StartOfLine);
351   // Adjust Lines to only contain relevant text.
352   Lines[LineIndex - 1] = Lines[LineIndex - 1].substr(0, EndOfPreviousLine);
353   Lines[LineIndex] = Lines[LineIndex].substr(StartOfLine);
354   // Adjust LeadingWhitespace to account all whitespace between the lines
355   // to the current line.
356   LeadingWhitespace[LineIndex] =
357       Lines[LineIndex].begin() - Lines[LineIndex - 1].end();
358 
359   // Adjust the start column uniformly across all lines.
360   StartOfLineColumn[LineIndex] =
361       encoding::columnWidthWithTabs(Whitespace, 0, Style.TabWidth, Encoding) +
362       IndentDelta;
363 }
364 
365 unsigned BreakableBlockComment::getLineCount() const { return Lines.size(); }
366 
367 unsigned BreakableBlockComment::getLineLengthAfterSplit(
368     unsigned LineIndex, unsigned Offset, StringRef::size_type Length) const {
369   unsigned ContentStartColumn = getContentStartColumn(LineIndex, Offset);
370   return ContentStartColumn +
371          encoding::columnWidthWithTabs(Lines[LineIndex].substr(Offset, Length),
372                                        ContentStartColumn, Style.TabWidth,
373                                        Encoding) +
374          // The last line gets a "*/" postfix.
375          (LineIndex + 1 == Lines.size() ? 2 : 0);
376 }
377 
378 BreakableToken::Split
379 BreakableBlockComment::getSplit(unsigned LineIndex, unsigned TailOffset,
380                                 unsigned ColumnLimit) const {
381   return getCommentSplit(Lines[LineIndex].substr(TailOffset),
382                          getContentStartColumn(LineIndex, TailOffset),
383                          ColumnLimit, Style.TabWidth, Encoding);
384 }
385 
386 void BreakableBlockComment::insertBreak(unsigned LineIndex, unsigned TailOffset,
387                                         Split Split,
388                                         WhitespaceManager &Whitespaces) {
389   StringRef Text = Lines[LineIndex].substr(TailOffset);
390   StringRef Prefix = Decoration;
391   if (LineIndex + 1 == Lines.size() &&
392       Text.size() == Split.first + Split.second) {
393     // For the last line we need to break before "*/", but not to add "* ".
394     Prefix = "";
395   }
396 
397   unsigned BreakOffsetInToken =
398       Text.data() - Tok.TokenText.data() + Split.first;
399   unsigned CharsToRemove = Split.second;
400   assert(IndentAtLineBreak >= Decoration.size());
401   Whitespaces.replaceWhitespaceInToken(
402       Tok, BreakOffsetInToken, CharsToRemove, "", Prefix, InPPDirective, 1,
403       IndentLevel, IndentAtLineBreak - Decoration.size());
404 }
405 
406 void BreakableBlockComment::replaceWhitespace(unsigned LineIndex,
407                                               unsigned TailOffset, Split Split,
408                                               WhitespaceManager &Whitespaces) {
409   StringRef Text = Lines[LineIndex].substr(TailOffset);
410   unsigned BreakOffsetInToken =
411       Text.data() - Tok.TokenText.data() + Split.first;
412   unsigned CharsToRemove = Split.second;
413   Whitespaces.replaceWhitespaceInToken(
414       Tok, BreakOffsetInToken, CharsToRemove, "", "", /*InPPDirective=*/false,
415       /*Newlines=*/0, /*IndentLevel=*/0, /*Spaces=*/1);
416 }
417 
418 void
419 BreakableBlockComment::replaceWhitespaceBefore(unsigned LineIndex,
420                                                WhitespaceManager &Whitespaces) {
421   if (LineIndex == 0)
422     return;
423   StringRef Prefix = Decoration;
424   if (Lines[LineIndex].empty()) {
425     if (LineIndex + 1 == Lines.size()) {
426       if (!LastLineNeedsDecoration) {
427         // If the last line was empty, we don't need a prefix, as the */ will
428         // line up with the decoration (if it exists).
429         Prefix = "";
430       }
431     } else if (!Decoration.empty()) {
432       // For other empty lines, if we do have a decoration, adapt it to not
433       // contain a trailing whitespace.
434       Prefix = Prefix.substr(0, 1);
435     }
436   } else {
437     if (StartOfLineColumn[LineIndex] == 1) {
438       // This line starts immediately after the decorating *.
439       Prefix = Prefix.substr(0, 1);
440     }
441   }
442 
443   unsigned WhitespaceOffsetInToken = Lines[LineIndex].data() -
444                                      Tok.TokenText.data() -
445                                      LeadingWhitespace[LineIndex];
446   Whitespaces.replaceWhitespaceInToken(
447       Tok, WhitespaceOffsetInToken, LeadingWhitespace[LineIndex], "", Prefix,
448       InPPDirective, 1, IndentLevel,
449       StartOfLineColumn[LineIndex] - Prefix.size());
450 }
451 
452 unsigned
453 BreakableBlockComment::getContentStartColumn(unsigned LineIndex,
454                                              unsigned TailOffset) const {
455   // If we break, we always break at the predefined indent.
456   if (TailOffset != 0)
457     return IndentAtLineBreak;
458   return std::max(0, StartOfLineColumn[LineIndex]);
459 }
460 
461 } // namespace format
462 } // namespace clang
463