1//===--- Rewriter.cpp - Code rewriting interface --------------------------===//
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//  This file defines the Rewriter class, which is used for code
11//  transformations.
12//
13//===----------------------------------------------------------------------===//
14
15#include "clang/Rewrite/Core/Rewriter.h"
16#include "clang/AST/Stmt.h"
17#include "clang/AST/Decl.h"
18#include "clang/Basic/DiagnosticIDs.h"
19#include "clang/Basic/FileManager.h"
20#include "clang/Basic/SourceManager.h"
21#include "clang/Lex/Lexer.h"
22#include "llvm/ADT/SmallString.h"
23#include "llvm/Support/FileSystem.h"
24using namespace clang;
25
26raw_ostream &RewriteBuffer::write(raw_ostream &os) const {
27  // FIXME: eliminate the copy by writing out each chunk at a time
28  os << std::string(begin(), end());
29  return os;
30}
31
32/// \brief Return true if this character is non-new-line whitespace:
33/// ' ', '\\t', '\\f', '\\v', '\\r'.
34static inline bool isWhitespace(unsigned char c) {
35  switch (c) {
36  case ' ':
37  case '\t':
38  case '\f':
39  case '\v':
40  case '\r':
41    return true;
42  default:
43    return false;
44  }
45}
46
47void RewriteBuffer::RemoveText(unsigned OrigOffset, unsigned Size,
48                               bool removeLineIfEmpty) {
49  // Nothing to remove, exit early.
50  if (Size == 0) return;
51
52  unsigned RealOffset = getMappedOffset(OrigOffset, true);
53  assert(RealOffset+Size < Buffer.size() && "Invalid location");
54
55  // Remove the dead characters.
56  Buffer.erase(RealOffset, Size);
57
58  // Add a delta so that future changes are offset correctly.
59  AddReplaceDelta(OrigOffset, -Size);
60
61  if (removeLineIfEmpty) {
62    // Find the line that the remove occurred and if it is completely empty
63    // remove the line as well.
64
65    iterator curLineStart = begin();
66    unsigned curLineStartOffs = 0;
67    iterator posI = begin();
68    for (unsigned i = 0; i != RealOffset; ++i) {
69      if (*posI == '\n') {
70        curLineStart = posI;
71        ++curLineStart;
72        curLineStartOffs = i + 1;
73      }
74      ++posI;
75    }
76
77    unsigned lineSize = 0;
78    posI = curLineStart;
79    while (posI != end() && isWhitespace(*posI)) {
80      ++posI;
81      ++lineSize;
82    }
83    if (posI != end() && *posI == '\n') {
84      Buffer.erase(curLineStartOffs, lineSize + 1/* + '\n'*/);
85      AddReplaceDelta(curLineStartOffs, -(lineSize + 1/* + '\n'*/));
86    }
87  }
88}
89
90void RewriteBuffer::InsertText(unsigned OrigOffset, StringRef Str,
91                               bool InsertAfter) {
92
93  // Nothing to insert, exit early.
94  if (Str.empty()) return;
95
96  unsigned RealOffset = getMappedOffset(OrigOffset, InsertAfter);
97  Buffer.insert(RealOffset, Str.begin(), Str.end());
98
99  // Add a delta so that future changes are offset correctly.
100  AddInsertDelta(OrigOffset, Str.size());
101}
102
103/// ReplaceText - This method replaces a range of characters in the input
104/// buffer with a new string.  This is effectively a combined "remove+insert"
105/// operation.
106void RewriteBuffer::ReplaceText(unsigned OrigOffset, unsigned OrigLength,
107                                StringRef NewStr) {
108  unsigned RealOffset = getMappedOffset(OrigOffset, true);
109  Buffer.erase(RealOffset, OrigLength);
110  Buffer.insert(RealOffset, NewStr.begin(), NewStr.end());
111  if (OrigLength != NewStr.size())
112    AddReplaceDelta(OrigOffset, NewStr.size() - OrigLength);
113}
114
115
116//===----------------------------------------------------------------------===//
117// Rewriter class
118//===----------------------------------------------------------------------===//
119
120/// getRangeSize - Return the size in bytes of the specified range if they
121/// are in the same file.  If not, this returns -1.
122int Rewriter::getRangeSize(const CharSourceRange &Range,
123                           RewriteOptions opts) const {
124  if (!isRewritable(Range.getBegin()) ||
125      !isRewritable(Range.getEnd())) return -1;
126
127  FileID StartFileID, EndFileID;
128  unsigned StartOff, EndOff;
129
130  StartOff = getLocationOffsetAndFileID(Range.getBegin(), StartFileID);
131  EndOff   = getLocationOffsetAndFileID(Range.getEnd(), EndFileID);
132
133  if (StartFileID != EndFileID)
134    return -1;
135
136  // If edits have been made to this buffer, the delta between the range may
137  // have changed.
138  std::map<FileID, RewriteBuffer>::const_iterator I =
139    RewriteBuffers.find(StartFileID);
140  if (I != RewriteBuffers.end()) {
141    const RewriteBuffer &RB = I->second;
142    EndOff = RB.getMappedOffset(EndOff, opts.IncludeInsertsAtEndOfRange);
143    StartOff = RB.getMappedOffset(StartOff, !opts.IncludeInsertsAtBeginOfRange);
144  }
145
146
147  // Adjust the end offset to the end of the last token, instead of being the
148  // start of the last token if this is a token range.
149  if (Range.isTokenRange())
150    EndOff += Lexer::MeasureTokenLength(Range.getEnd(), *SourceMgr, *LangOpts);
151
152  return EndOff-StartOff;
153}
154
155int Rewriter::getRangeSize(SourceRange Range, RewriteOptions opts) const {
156  return getRangeSize(CharSourceRange::getTokenRange(Range), opts);
157}
158
159
160/// getRewrittenText - Return the rewritten form of the text in the specified
161/// range.  If the start or end of the range was unrewritable or if they are
162/// in different buffers, this returns an empty string.
163///
164/// Note that this method is not particularly efficient.
165///
166std::string Rewriter::getRewrittenText(SourceRange Range) const {
167  if (!isRewritable(Range.getBegin()) ||
168      !isRewritable(Range.getEnd()))
169    return "";
170
171  FileID StartFileID, EndFileID;
172  unsigned StartOff, EndOff;
173  StartOff = getLocationOffsetAndFileID(Range.getBegin(), StartFileID);
174  EndOff   = getLocationOffsetAndFileID(Range.getEnd(), EndFileID);
175
176  if (StartFileID != EndFileID)
177    return ""; // Start and end in different buffers.
178
179  // If edits have been made to this buffer, the delta between the range may
180  // have changed.
181  std::map<FileID, RewriteBuffer>::const_iterator I =
182    RewriteBuffers.find(StartFileID);
183  if (I == RewriteBuffers.end()) {
184    // If the buffer hasn't been rewritten, just return the text from the input.
185    const char *Ptr = SourceMgr->getCharacterData(Range.getBegin());
186
187    // Adjust the end offset to the end of the last token, instead of being the
188    // start of the last token.
189    EndOff += Lexer::MeasureTokenLength(Range.getEnd(), *SourceMgr, *LangOpts);
190    return std::string(Ptr, Ptr+EndOff-StartOff);
191  }
192
193  const RewriteBuffer &RB = I->second;
194  EndOff = RB.getMappedOffset(EndOff, true);
195  StartOff = RB.getMappedOffset(StartOff);
196
197  // Adjust the end offset to the end of the last token, instead of being the
198  // start of the last token.
199  EndOff += Lexer::MeasureTokenLength(Range.getEnd(), *SourceMgr, *LangOpts);
200
201  // Advance the iterators to the right spot, yay for linear time algorithms.
202  RewriteBuffer::iterator Start = RB.begin();
203  std::advance(Start, StartOff);
204  RewriteBuffer::iterator End = Start;
205  std::advance(End, EndOff-StartOff);
206
207  return std::string(Start, End);
208}
209
210unsigned Rewriter::getLocationOffsetAndFileID(SourceLocation Loc,
211                                              FileID &FID) const {
212  assert(Loc.isValid() && "Invalid location");
213  std::pair<FileID,unsigned> V = SourceMgr->getDecomposedLoc(Loc);
214  FID = V.first;
215  return V.second;
216}
217
218
219/// getEditBuffer - Get or create a RewriteBuffer for the specified FileID.
220///
221RewriteBuffer &Rewriter::getEditBuffer(FileID FID) {
222  std::map<FileID, RewriteBuffer>::iterator I =
223    RewriteBuffers.lower_bound(FID);
224  if (I != RewriteBuffers.end() && I->first == FID)
225    return I->second;
226  I = RewriteBuffers.insert(I, std::make_pair(FID, RewriteBuffer()));
227
228  StringRef MB = SourceMgr->getBufferData(FID);
229  I->second.Initialize(MB.begin(), MB.end());
230
231  return I->second;
232}
233
234/// InsertText - Insert the specified string at the specified location in the
235/// original buffer.
236bool Rewriter::InsertText(SourceLocation Loc, StringRef Str,
237                          bool InsertAfter, bool indentNewLines) {
238  if (!isRewritable(Loc)) return true;
239  FileID FID;
240  unsigned StartOffs = getLocationOffsetAndFileID(Loc, FID);
241
242  SmallString<128> indentedStr;
243  if (indentNewLines && Str.find('\n') != StringRef::npos) {
244    StringRef MB = SourceMgr->getBufferData(FID);
245
246    unsigned lineNo = SourceMgr->getLineNumber(FID, StartOffs) - 1;
247    const SrcMgr::ContentCache *
248        Content = SourceMgr->getSLocEntry(FID).getFile().getContentCache();
249    unsigned lineOffs = Content->SourceLineCache[lineNo];
250
251    // Find the whitespace at the start of the line.
252    StringRef indentSpace;
253    {
254      unsigned i = lineOffs;
255      while (isWhitespace(MB[i]))
256        ++i;
257      indentSpace = MB.substr(lineOffs, i-lineOffs);
258    }
259
260    SmallVector<StringRef, 4> lines;
261    Str.split(lines, "\n");
262
263    for (unsigned i = 0, e = lines.size(); i != e; ++i) {
264      indentedStr += lines[i];
265      if (i < e-1) {
266        indentedStr += '\n';
267        indentedStr += indentSpace;
268      }
269    }
270    Str = indentedStr.str();
271  }
272
273  getEditBuffer(FID).InsertText(StartOffs, Str, InsertAfter);
274  return false;
275}
276
277bool Rewriter::InsertTextAfterToken(SourceLocation Loc, StringRef Str) {
278  if (!isRewritable(Loc)) return true;
279  FileID FID;
280  unsigned StartOffs = getLocationOffsetAndFileID(Loc, FID);
281  RewriteOptions rangeOpts;
282  rangeOpts.IncludeInsertsAtBeginOfRange = false;
283  StartOffs += getRangeSize(SourceRange(Loc, Loc), rangeOpts);
284  getEditBuffer(FID).InsertText(StartOffs, Str, /*InsertAfter*/true);
285  return false;
286}
287
288/// RemoveText - Remove the specified text region.
289bool Rewriter::RemoveText(SourceLocation Start, unsigned Length,
290                          RewriteOptions opts) {
291  if (!isRewritable(Start)) return true;
292  FileID FID;
293  unsigned StartOffs = getLocationOffsetAndFileID(Start, FID);
294  getEditBuffer(FID).RemoveText(StartOffs, Length, opts.RemoveLineIfEmpty);
295  return false;
296}
297
298/// ReplaceText - This method replaces a range of characters in the input
299/// buffer with a new string.  This is effectively a combined "remove/insert"
300/// operation.
301bool Rewriter::ReplaceText(SourceLocation Start, unsigned OrigLength,
302                           StringRef NewStr) {
303  if (!isRewritable(Start)) return true;
304  FileID StartFileID;
305  unsigned StartOffs = getLocationOffsetAndFileID(Start, StartFileID);
306
307  getEditBuffer(StartFileID).ReplaceText(StartOffs, OrigLength, NewStr);
308  return false;
309}
310
311bool Rewriter::ReplaceText(SourceRange range, SourceRange replacementRange) {
312  if (!isRewritable(range.getBegin())) return true;
313  if (!isRewritable(range.getEnd())) return true;
314  if (replacementRange.isInvalid()) return true;
315  SourceLocation start = range.getBegin();
316  unsigned origLength = getRangeSize(range);
317  unsigned newLength = getRangeSize(replacementRange);
318  FileID FID;
319  unsigned newOffs = getLocationOffsetAndFileID(replacementRange.getBegin(),
320                                                FID);
321  StringRef MB = SourceMgr->getBufferData(FID);
322  return ReplaceText(start, origLength, MB.substr(newOffs, newLength));
323}
324
325/// ReplaceStmt - This replaces a Stmt/Expr with another, using the pretty
326/// printer to generate the replacement code.  This returns true if the input
327/// could not be rewritten, or false if successful.
328bool Rewriter::ReplaceStmt(Stmt *From, Stmt *To) {
329  // Measaure the old text.
330  int Size = getRangeSize(From->getSourceRange());
331  if (Size == -1)
332    return true;
333
334  // Get the new text.
335  std::string SStr;
336  llvm::raw_string_ostream S(SStr);
337  To->printPretty(S, 0, PrintingPolicy(*LangOpts));
338  const std::string &Str = S.str();
339
340  ReplaceText(From->getLocStart(), Size, Str);
341  return false;
342}
343
344std::string Rewriter::ConvertToString(Stmt *From) {
345  std::string SStr;
346  llvm::raw_string_ostream S(SStr);
347  From->printPretty(S, 0, PrintingPolicy(*LangOpts));
348  return S.str();
349}
350
351bool Rewriter::IncreaseIndentation(CharSourceRange range,
352                                   SourceLocation parentIndent) {
353  if (range.isInvalid()) return true;
354  if (!isRewritable(range.getBegin())) return true;
355  if (!isRewritable(range.getEnd())) return true;
356  if (!isRewritable(parentIndent)) return true;
357
358  FileID StartFileID, EndFileID, parentFileID;
359  unsigned StartOff, EndOff, parentOff;
360
361  StartOff = getLocationOffsetAndFileID(range.getBegin(), StartFileID);
362  EndOff   = getLocationOffsetAndFileID(range.getEnd(), EndFileID);
363  parentOff = getLocationOffsetAndFileID(parentIndent, parentFileID);
364
365  if (StartFileID != EndFileID || StartFileID != parentFileID)
366    return true;
367  if (StartOff > EndOff)
368    return true;
369
370  FileID FID = StartFileID;
371  StringRef MB = SourceMgr->getBufferData(FID);
372
373  unsigned parentLineNo = SourceMgr->getLineNumber(FID, parentOff) - 1;
374  unsigned startLineNo = SourceMgr->getLineNumber(FID, StartOff) - 1;
375  unsigned endLineNo = SourceMgr->getLineNumber(FID, EndOff) - 1;
376
377  const SrcMgr::ContentCache *
378      Content = SourceMgr->getSLocEntry(FID).getFile().getContentCache();
379
380  // Find where the lines start.
381  unsigned parentLineOffs = Content->SourceLineCache[parentLineNo];
382  unsigned startLineOffs = Content->SourceLineCache[startLineNo];
383
384  // Find the whitespace at the start of each line.
385  StringRef parentSpace, startSpace;
386  {
387    unsigned i = parentLineOffs;
388    while (isWhitespace(MB[i]))
389      ++i;
390    parentSpace = MB.substr(parentLineOffs, i-parentLineOffs);
391
392    i = startLineOffs;
393    while (isWhitespace(MB[i]))
394      ++i;
395    startSpace = MB.substr(startLineOffs, i-startLineOffs);
396  }
397  if (parentSpace.size() >= startSpace.size())
398    return true;
399  if (!startSpace.startswith(parentSpace))
400    return true;
401
402  StringRef indent = startSpace.substr(parentSpace.size());
403
404  // Indent the lines between start/end offsets.
405  RewriteBuffer &RB = getEditBuffer(FID);
406  for (unsigned lineNo = startLineNo; lineNo <= endLineNo; ++lineNo) {
407    unsigned offs = Content->SourceLineCache[lineNo];
408    unsigned i = offs;
409    while (isWhitespace(MB[i]))
410      ++i;
411    StringRef origIndent = MB.substr(offs, i-offs);
412    if (origIndent.startswith(startSpace))
413      RB.InsertText(offs, indent, /*InsertAfter=*/false);
414  }
415
416  return false;
417}
418
419// A wrapper for a file stream that atomically overwrites the target.
420//
421// Creates a file output stream for a temporary file in the constructor,
422// which is later accessible via getStream() if ok() return true.
423// Flushes the stream and moves the temporary file to the target location
424// in the destructor.
425class AtomicallyMovedFile {
426public:
427  AtomicallyMovedFile(DiagnosticsEngine &Diagnostics, StringRef Filename,
428                      bool &AllWritten)
429    : Diagnostics(Diagnostics), Filename(Filename), AllWritten(AllWritten) {
430    TempFilename = Filename;
431    TempFilename += "-%%%%%%%%";
432    int FD;
433    if (llvm::sys::fs::unique_file(TempFilename.str(), FD, TempFilename,
434                                    /*makeAbsolute=*/true, 0664)) {
435      AllWritten = false;
436      Diagnostics.Report(clang::diag::err_unable_to_make_temp)
437        << TempFilename;
438    } else {
439      FileStream.reset(new llvm::raw_fd_ostream(FD, /*shouldClose=*/true));
440    }
441  }
442
443  ~AtomicallyMovedFile() {
444    if (!ok()) return;
445
446    FileStream->flush();
447#ifdef _WIN32
448    // Win32 does not allow rename/removing opened files.
449    FileStream.reset();
450#endif
451    if (llvm::error_code ec =
452          llvm::sys::fs::rename(TempFilename.str(), Filename)) {
453      AllWritten = false;
454      Diagnostics.Report(clang::diag::err_unable_to_rename_temp)
455        << TempFilename << Filename << ec.message();
456      bool existed;
457      // If the remove fails, there's not a lot we can do - this is already an
458      // error.
459      llvm::sys::fs::remove(TempFilename.str(), existed);
460    }
461  }
462
463  bool ok() { return FileStream; }
464  llvm::raw_ostream &getStream() { return *FileStream; }
465
466private:
467  DiagnosticsEngine &Diagnostics;
468  StringRef Filename;
469  SmallString<128> TempFilename;
470  OwningPtr<llvm::raw_fd_ostream> FileStream;
471  bool &AllWritten;
472};
473
474bool Rewriter::overwriteChangedFiles() {
475  bool AllWritten = true;
476  for (buffer_iterator I = buffer_begin(), E = buffer_end(); I != E; ++I) {
477    const FileEntry *Entry =
478        getSourceMgr().getFileEntryForID(I->first);
479    AtomicallyMovedFile File(getSourceMgr().getDiagnostics(), Entry->getName(),
480                             AllWritten);
481    if (File.ok()) {
482      I->second.write(File.getStream());
483    }
484  }
485  return !AllWritten;
486}
487