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
25namespace clang {
26namespace format {
27
28static const char *const Blanks = " \t\v\f\r";
29static 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
42static 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
85static BreakableToken::Split
86getStringSplit(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
134unsigned BreakableSingleLineToken::getLineCount() const { return 1; }
135
136unsigned 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
144BreakableSingleLineToken::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
155BreakableStringLiteral::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
162BreakableToken::Split
163BreakableStringLiteral::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
170void 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
185static 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
200BreakableLineComment::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  }
214}
215
216BreakableToken::Split
217BreakableLineComment::getSplit(unsigned LineIndex, unsigned TailOffset,
218                               unsigned ColumnLimit) const {
219  return getCommentSplit(Line.substr(TailOffset), StartColumn + Prefix.size(),
220                         ColumnLimit, Style.TabWidth, Encoding);
221}
222
223void BreakableLineComment::insertBreak(unsigned LineIndex, unsigned TailOffset,
224                                       Split Split,
225                                       WhitespaceManager &Whitespaces) {
226  Whitespaces.replaceWhitespaceInToken(
227      Tok, OriginalPrefix.size() + TailOffset + Split.first, Split.second,
228      Postfix, Prefix, InPPDirective, /*Newlines=*/1, IndentLevel, StartColumn);
229}
230
231void BreakableLineComment::replaceWhitespace(unsigned LineIndex,
232                                             unsigned TailOffset, Split Split,
233                                             WhitespaceManager &Whitespaces) {
234  Whitespaces.replaceWhitespaceInToken(
235      Tok, OriginalPrefix.size() + TailOffset + Split.first, Split.second, "",
236      "", /*InPPDirective=*/false, /*Newlines=*/0, /*IndentLevel=*/0,
237      /*Spaces=*/1);
238}
239
240void
241BreakableLineComment::replaceWhitespaceBefore(unsigned LineIndex,
242                                              WhitespaceManager &Whitespaces) {
243  if (OriginalPrefix != Prefix) {
244    Whitespaces.replaceWhitespaceInToken(Tok, OriginalPrefix.size(), 0, "", "",
245                                         /*InPPDirective=*/false,
246                                         /*Newlines=*/0, /*IndentLevel=*/0,
247                                         /*Spaces=*/1);
248  }
249}
250
251BreakableBlockComment::BreakableBlockComment(
252    const FormatToken &Token, unsigned IndentLevel, unsigned StartColumn,
253    unsigned OriginalStartColumn, bool FirstInLine, bool InPPDirective,
254    encoding::Encoding Encoding, const FormatStyle &Style)
255    : BreakableToken(Token, IndentLevel, InPPDirective, Encoding, Style) {
256  StringRef TokenText(Token.TokenText);
257  assert(TokenText.startswith("/*") && TokenText.endswith("*/"));
258  TokenText.substr(2, TokenText.size() - 4).split(Lines, "\n");
259
260  int IndentDelta = StartColumn - OriginalStartColumn;
261  LeadingWhitespace.resize(Lines.size());
262  StartOfLineColumn.resize(Lines.size());
263  StartOfLineColumn[0] = StartColumn + 2;
264  for (size_t i = 1; i < Lines.size(); ++i)
265    adjustWhitespace(i, IndentDelta);
266
267  Decoration = "* ";
268  if (Lines.size() == 1 && !FirstInLine) {
269    // Comments for which FirstInLine is false can start on arbitrary column,
270    // and available horizontal space can be too small to align consecutive
271    // lines with the first one.
272    // FIXME: We could, probably, align them to current indentation level, but
273    // now we just wrap them without stars.
274    Decoration = "";
275  }
276  for (size_t i = 1, e = Lines.size(); i < e && !Decoration.empty(); ++i) {
277    // If the last line is empty, the closing "*/" will have a star.
278    if (i + 1 == e && Lines[i].empty())
279      break;
280    while (!Lines[i].startswith(Decoration))
281      Decoration = Decoration.substr(0, Decoration.size() - 1);
282  }
283
284  LastLineNeedsDecoration = true;
285  IndentAtLineBreak = StartOfLineColumn[0] + 1;
286  for (size_t i = 1; i < Lines.size(); ++i) {
287    if (Lines[i].empty()) {
288      if (i + 1 == Lines.size()) {
289        // Empty last line means that we already have a star as a part of the
290        // trailing */. We also need to preserve whitespace, so that */ is
291        // correctly indented.
292        LastLineNeedsDecoration = false;
293      } else if (Decoration.empty()) {
294        // For all other lines, set the start column to 0 if they're empty, so
295        // we do not insert trailing whitespace anywhere.
296        StartOfLineColumn[i] = 0;
297      }
298      continue;
299    }
300    // The first line already excludes the star.
301    // For all other lines, adjust the line to exclude the star and
302    // (optionally) the first whitespace.
303    StartOfLineColumn[i] += Decoration.size();
304    Lines[i] = Lines[i].substr(Decoration.size());
305    LeadingWhitespace[i] += Decoration.size();
306    IndentAtLineBreak = std::min<int>(IndentAtLineBreak, StartOfLineColumn[i]);
307  }
308  IndentAtLineBreak = std::max<unsigned>(IndentAtLineBreak, Decoration.size());
309  DEBUG({
310    llvm::dbgs() << "IndentAtLineBreak " << IndentAtLineBreak << "\n";
311    for (size_t i = 0; i < Lines.size(); ++i) {
312      llvm::dbgs() << i << " |" << Lines[i] << "| " << LeadingWhitespace[i]
313                   << "\n";
314    }
315  });
316}
317
318void BreakableBlockComment::adjustWhitespace(unsigned LineIndex,
319                                             int IndentDelta) {
320  // When in a preprocessor directive, the trailing backslash in a block comment
321  // is not needed, but can serve a purpose of uniformity with necessary escaped
322  // newlines outside the comment. In this case we remove it here before
323  // trimming the trailing whitespace. The backslash will be re-added later when
324  // inserting a line break.
325  size_t EndOfPreviousLine = Lines[LineIndex - 1].size();
326  if (InPPDirective && Lines[LineIndex - 1].endswith("\\"))
327    --EndOfPreviousLine;
328
329  // Calculate the end of the non-whitespace text in the previous line.
330  EndOfPreviousLine =
331      Lines[LineIndex - 1].find_last_not_of(Blanks, EndOfPreviousLine);
332  if (EndOfPreviousLine == StringRef::npos)
333    EndOfPreviousLine = 0;
334  else
335    ++EndOfPreviousLine;
336  // Calculate the start of the non-whitespace text in the current line.
337  size_t StartOfLine = Lines[LineIndex].find_first_not_of(Blanks);
338  if (StartOfLine == StringRef::npos)
339    StartOfLine = Lines[LineIndex].size();
340
341  StringRef Whitespace = Lines[LineIndex].substr(0, StartOfLine);
342  // Adjust Lines to only contain relevant text.
343  Lines[LineIndex - 1] = Lines[LineIndex - 1].substr(0, EndOfPreviousLine);
344  Lines[LineIndex] = Lines[LineIndex].substr(StartOfLine);
345  // Adjust LeadingWhitespace to account all whitespace between the lines
346  // to the current line.
347  LeadingWhitespace[LineIndex] =
348      Lines[LineIndex].begin() - Lines[LineIndex - 1].end();
349
350  // Adjust the start column uniformly across all lines.
351  StartOfLineColumn[LineIndex] =
352      encoding::columnWidthWithTabs(Whitespace, 0, Style.TabWidth, Encoding) +
353      IndentDelta;
354}
355
356unsigned BreakableBlockComment::getLineCount() const { return Lines.size(); }
357
358unsigned BreakableBlockComment::getLineLengthAfterSplit(
359    unsigned LineIndex, unsigned Offset, StringRef::size_type Length) const {
360  unsigned ContentStartColumn = getContentStartColumn(LineIndex, Offset);
361  return ContentStartColumn +
362         encoding::columnWidthWithTabs(Lines[LineIndex].substr(Offset, Length),
363                                       ContentStartColumn, Style.TabWidth,
364                                       Encoding) +
365         // The last line gets a "*/" postfix.
366         (LineIndex + 1 == Lines.size() ? 2 : 0);
367}
368
369BreakableToken::Split
370BreakableBlockComment::getSplit(unsigned LineIndex, unsigned TailOffset,
371                                unsigned ColumnLimit) const {
372  return getCommentSplit(Lines[LineIndex].substr(TailOffset),
373                         getContentStartColumn(LineIndex, TailOffset),
374                         ColumnLimit, Style.TabWidth, Encoding);
375}
376
377void BreakableBlockComment::insertBreak(unsigned LineIndex, unsigned TailOffset,
378                                        Split Split,
379                                        WhitespaceManager &Whitespaces) {
380  StringRef Text = Lines[LineIndex].substr(TailOffset);
381  StringRef Prefix = Decoration;
382  if (LineIndex + 1 == Lines.size() &&
383      Text.size() == Split.first + Split.second) {
384    // For the last line we need to break before "*/", but not to add "* ".
385    Prefix = "";
386  }
387
388  unsigned BreakOffsetInToken =
389      Text.data() - Tok.TokenText.data() + Split.first;
390  unsigned CharsToRemove = Split.second;
391  assert(IndentAtLineBreak >= Decoration.size());
392  Whitespaces.replaceWhitespaceInToken(
393      Tok, BreakOffsetInToken, CharsToRemove, "", Prefix, InPPDirective, 1,
394      IndentLevel, IndentAtLineBreak - Decoration.size());
395}
396
397void BreakableBlockComment::replaceWhitespace(unsigned LineIndex,
398                                              unsigned TailOffset, Split Split,
399                                              WhitespaceManager &Whitespaces) {
400  StringRef Text = Lines[LineIndex].substr(TailOffset);
401  unsigned BreakOffsetInToken =
402      Text.data() - Tok.TokenText.data() + Split.first;
403  unsigned CharsToRemove = Split.second;
404  Whitespaces.replaceWhitespaceInToken(
405      Tok, BreakOffsetInToken, CharsToRemove, "", "", /*InPPDirective=*/false,
406      /*Newlines=*/0, /*IndentLevel=*/0, /*Spaces=*/1);
407}
408
409void
410BreakableBlockComment::replaceWhitespaceBefore(unsigned LineIndex,
411                                               WhitespaceManager &Whitespaces) {
412  if (LineIndex == 0)
413    return;
414  StringRef Prefix = Decoration;
415  if (Lines[LineIndex].empty()) {
416    if (LineIndex + 1 == Lines.size()) {
417      if (!LastLineNeedsDecoration) {
418        // If the last line was empty, we don't need a prefix, as the */ will
419        // line up with the decoration (if it exists).
420        Prefix = "";
421      }
422    } else if (!Decoration.empty()) {
423      // For other empty lines, if we do have a decoration, adapt it to not
424      // contain a trailing whitespace.
425      Prefix = Prefix.substr(0, 1);
426    }
427  } else {
428    if (StartOfLineColumn[LineIndex] == 1) {
429      // This line starts immediately after the decorating *.
430      Prefix = Prefix.substr(0, 1);
431    }
432  }
433
434  unsigned WhitespaceOffsetInToken = Lines[LineIndex].data() -
435                                     Tok.TokenText.data() -
436                                     LeadingWhitespace[LineIndex];
437  Whitespaces.replaceWhitespaceInToken(
438      Tok, WhitespaceOffsetInToken, LeadingWhitespace[LineIndex], "", Prefix,
439      InPPDirective, 1, IndentLevel,
440      StartOfLineColumn[LineIndex] - Prefix.size());
441}
442
443unsigned
444BreakableBlockComment::getContentStartColumn(unsigned LineIndex,
445                                             unsigned TailOffset) const {
446  // If we break, we always break at the predefined indent.
447  if (TailOffset != 0)
448    return IndentAtLineBreak;
449  return std::max(0, StartOfLineColumn[LineIndex]);
450}
451
452} // namespace format
453} // namespace clang
454