1 //===--- RawCommentList.cpp - Processing raw comments -----------*- C++ -*-===// 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 #include "clang/AST/RawCommentList.h" 11 #include "clang/AST/ASTContext.h" 12 #include "clang/AST/CommentLexer.h" 13 #include "clang/AST/CommentBriefParser.h" 14 #include "llvm/ADT/STLExtras.h" 15 16 using namespace clang; 17 18 namespace { 19 /// Get comment kind and bool describing if it is a trailing comment. 20 std::pair<RawComment::CommentKind, bool> getCommentKind(StringRef Comment) { 21 if (Comment.size() < 3 || Comment[0] != '/') 22 return std::make_pair(RawComment::CK_Invalid, false); 23 24 RawComment::CommentKind K; 25 if (Comment[1] == '/') { 26 if (Comment.size() < 3) 27 return std::make_pair(RawComment::CK_OrdinaryBCPL, false); 28 29 if (Comment[2] == '/') 30 K = RawComment::CK_BCPLSlash; 31 else if (Comment[2] == '!') 32 K = RawComment::CK_BCPLExcl; 33 else 34 return std::make_pair(RawComment::CK_OrdinaryBCPL, false); 35 } else { 36 assert(Comment.size() >= 4); 37 38 // Comment lexer does not understand escapes in comment markers, so pretend 39 // that this is not a comment. 40 if (Comment[1] != '*' || 41 Comment[Comment.size() - 2] != '*' || 42 Comment[Comment.size() - 1] != '/') 43 return std::make_pair(RawComment::CK_Invalid, false); 44 45 if (Comment[2] == '*') 46 K = RawComment::CK_JavaDoc; 47 else if (Comment[2] == '!') 48 K = RawComment::CK_Qt; 49 else 50 return std::make_pair(RawComment::CK_OrdinaryC, false); 51 } 52 const bool TrailingComment = (Comment.size() > 3) && (Comment[3] == '<'); 53 return std::make_pair(K, TrailingComment); 54 } 55 56 bool mergedCommentIsTrailingComment(StringRef Comment) { 57 return (Comment.size() > 3) && (Comment[3] == '<'); 58 } 59 } // unnamed namespace 60 61 RawComment::RawComment(const SourceManager &SourceMgr, SourceRange SR, 62 bool Merged) : 63 Range(SR), RawTextValid(false), BriefTextValid(false), 64 IsAlmostTrailingComment(false), 65 BeginLineValid(false), EndLineValid(false) { 66 // Extract raw comment text, if possible. 67 if (SR.getBegin() == SR.getEnd() || getRawText(SourceMgr).empty()) { 68 Kind = CK_Invalid; 69 return; 70 } 71 72 if (!Merged) { 73 // Guess comment kind. 74 std::pair<CommentKind, bool> K = getCommentKind(RawText); 75 Kind = K.first; 76 IsTrailingComment = K.second; 77 78 IsAlmostTrailingComment = RawText.startswith("//<") || 79 RawText.startswith("/*<"); 80 } else { 81 Kind = CK_Merged; 82 IsTrailingComment = mergedCommentIsTrailingComment(RawText); 83 } 84 } 85 86 unsigned RawComment::getBeginLine(const SourceManager &SM) const { 87 if (BeginLineValid) 88 return BeginLine; 89 90 std::pair<FileID, unsigned> LocInfo = SM.getDecomposedLoc(Range.getBegin()); 91 BeginLine = SM.getLineNumber(LocInfo.first, LocInfo.second); 92 BeginLineValid = true; 93 return BeginLine; 94 } 95 96 unsigned RawComment::getEndLine(const SourceManager &SM) const { 97 if (EndLineValid) 98 return EndLine; 99 100 std::pair<FileID, unsigned> LocInfo = SM.getDecomposedLoc(Range.getEnd()); 101 EndLine = SM.getLineNumber(LocInfo.first, LocInfo.second); 102 EndLineValid = true; 103 return EndLine; 104 } 105 106 StringRef RawComment::getRawTextSlow(const SourceManager &SourceMgr) const { 107 FileID BeginFileID; 108 FileID EndFileID; 109 unsigned BeginOffset; 110 unsigned EndOffset; 111 112 llvm::tie(BeginFileID, BeginOffset) = 113 SourceMgr.getDecomposedLoc(Range.getBegin()); 114 llvm::tie(EndFileID, EndOffset) = 115 SourceMgr.getDecomposedLoc(Range.getEnd()); 116 117 const unsigned Length = EndOffset - BeginOffset; 118 if (Length < 2) 119 return StringRef(); 120 121 // The comment can't begin in one file and end in another. 122 assert(BeginFileID == EndFileID); 123 124 bool Invalid = false; 125 const char *BufferStart = SourceMgr.getBufferData(BeginFileID, 126 &Invalid).data(); 127 if (Invalid) 128 return StringRef(); 129 130 return StringRef(BufferStart + BeginOffset, Length); 131 } 132 133 StringRef RawComment::extractBriefText(const ASTContext &Context) const { 134 // Make sure that RawText is valid. 135 getRawText(Context.getSourceManager()); 136 137 comments::Lexer L(Range.getBegin(), comments::CommentOptions(), 138 RawText.begin(), RawText.end()); 139 comments::BriefParser P(L); 140 141 const std::string Result = P.Parse(); 142 const unsigned BriefTextLength = Result.size(); 143 char *BriefTextPtr = new (Context) char[BriefTextLength + 1]; 144 memcpy(BriefTextPtr, Result.c_str(), BriefTextLength + 1); 145 BriefText = StringRef(BriefTextPtr, BriefTextLength); 146 BriefTextValid = true; 147 148 return BriefText; 149 } 150 151 namespace { 152 bool containsOnlyWhitespace(StringRef Str) { 153 return Str.find_first_not_of(" \t\f\v\r\n") == StringRef::npos; 154 } 155 156 bool onlyWhitespaceBetweenComments(SourceManager &SM, 157 const RawComment &C1, const RawComment &C2) { 158 std::pair<FileID, unsigned> C1EndLocInfo = SM.getDecomposedLoc( 159 C1.getSourceRange().getEnd()); 160 std::pair<FileID, unsigned> C2BeginLocInfo = SM.getDecomposedLoc( 161 C2.getSourceRange().getBegin()); 162 163 // Question does not make sense if comments are located in different files. 164 if (C1EndLocInfo.first != C2BeginLocInfo.first) 165 return false; 166 167 bool Invalid = false; 168 const char *Buffer = SM.getBufferData(C1EndLocInfo.first, &Invalid).data(); 169 if (Invalid) 170 return false; 171 172 StringRef TextBetweenComments(Buffer + C1EndLocInfo.second, 173 C2BeginLocInfo.second - C1EndLocInfo.second); 174 175 return containsOnlyWhitespace(TextBetweenComments); 176 } 177 } // unnamed namespace 178 179 void RawCommentList::addComment(const RawComment &RC) { 180 if (RC.isInvalid()) 181 return; 182 183 // Check if the comments are not in source order. 184 while (!Comments.empty() && 185 !SourceMgr.isBeforeInTranslationUnit( 186 Comments.back().getSourceRange().getBegin(), 187 RC.getSourceRange().getBegin())) { 188 // If they are, just pop a few last comments that don't fit. 189 // This happens if an \#include directive contains comments. 190 Comments.pop_back(); 191 } 192 193 if (OnlyWhitespaceSeen) { 194 if (!onlyWhitespaceBetweenComments(SourceMgr, LastComment, RC)) 195 OnlyWhitespaceSeen = false; 196 } 197 198 LastComment = RC; 199 200 // Ordinary comments are not interesting for us. 201 if (RC.isOrdinary()) 202 return; 203 204 // If this is the first Doxygen comment, save it (because there isn't 205 // anything to merge it with). 206 if (Comments.empty()) { 207 Comments.push_back(RC); 208 OnlyWhitespaceSeen = true; 209 return; 210 } 211 212 const RawComment &C1 = Comments.back(); 213 const RawComment &C2 = RC; 214 215 // Merge comments only if there is only whitespace between them. 216 // Can't merge trailing and non-trailing comments. 217 // Merge trailing comments if they are on same or consecutive lines. 218 if (OnlyWhitespaceSeen && 219 (C1.isTrailingComment() == C2.isTrailingComment()) && 220 (!C1.isTrailingComment() || 221 C1.getEndLine(SourceMgr) + 1 >= C2.getBeginLine(SourceMgr))) { 222 SourceRange MergedRange(C1.getSourceRange().getBegin(), 223 C2.getSourceRange().getEnd()); 224 RawComment Merged(SourceMgr, MergedRange, true); 225 Comments.pop_back(); 226 Comments.push_back(Merged); 227 } else 228 Comments.push_back(RC); 229 230 OnlyWhitespaceSeen = true; 231 } 232 233