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/Comment.h"
13 #include "clang/AST/CommentLexer.h"
14 #include "clang/AST/CommentBriefParser.h"
15 #include "clang/AST/CommentSema.h"
16 #include "clang/AST/CommentParser.h"
17 #include "clang/AST/CommentCommandTraits.h"
18 #include "llvm/ADT/STLExtras.h"
19 
20 using namespace clang;
21 
22 namespace {
23 /// Get comment kind and bool describing if it is a trailing comment.
24 std::pair<RawComment::CommentKind, bool> getCommentKind(StringRef Comment) {
25   if (Comment.size() < 3 || Comment[0] != '/')
26     return std::make_pair(RawComment::RCK_Invalid, false);
27 
28   RawComment::CommentKind K;
29   if (Comment[1] == '/') {
30     if (Comment.size() < 3)
31       return std::make_pair(RawComment::RCK_OrdinaryBCPL, false);
32 
33     if (Comment[2] == '/')
34       K = RawComment::RCK_BCPLSlash;
35     else if (Comment[2] == '!')
36       K = RawComment::RCK_BCPLExcl;
37     else
38       return std::make_pair(RawComment::RCK_OrdinaryBCPL, false);
39   } else {
40     assert(Comment.size() >= 4);
41 
42     // Comment lexer does not understand escapes in comment markers, so pretend
43     // that this is not a comment.
44     if (Comment[1] != '*' ||
45         Comment[Comment.size() - 2] != '*' ||
46         Comment[Comment.size() - 1] != '/')
47       return std::make_pair(RawComment::RCK_Invalid, false);
48 
49     if (Comment[2] == '*')
50       K = RawComment::RCK_JavaDoc;
51     else if (Comment[2] == '!')
52       K = RawComment::RCK_Qt;
53     else
54       return std::make_pair(RawComment::RCK_OrdinaryC, false);
55   }
56   const bool TrailingComment = (Comment.size() > 3) && (Comment[3] == '<');
57   return std::make_pair(K, TrailingComment);
58 }
59 
60 bool mergedCommentIsTrailingComment(StringRef Comment) {
61   return (Comment.size() > 3) && (Comment[3] == '<');
62 }
63 } // unnamed namespace
64 
65 RawComment::RawComment(const SourceManager &SourceMgr, SourceRange SR,
66                        bool Merged) :
67     Range(SR), RawTextValid(false), BriefTextValid(false),
68     IsAttached(false), IsAlmostTrailingComment(false),
69     BeginLineValid(false), EndLineValid(false) {
70   // Extract raw comment text, if possible.
71   if (SR.getBegin() == SR.getEnd() || getRawText(SourceMgr).empty()) {
72     Kind = RCK_Invalid;
73     return;
74   }
75 
76   if (!Merged) {
77     // Guess comment kind.
78     std::pair<CommentKind, bool> K = getCommentKind(RawText);
79     Kind = K.first;
80     IsTrailingComment = K.second;
81 
82     IsAlmostTrailingComment = RawText.startswith("//<") ||
83                                  RawText.startswith("/*<");
84   } else {
85     Kind = RCK_Merged;
86     IsTrailingComment = mergedCommentIsTrailingComment(RawText);
87   }
88 }
89 
90 unsigned RawComment::getBeginLine(const SourceManager &SM) const {
91   if (BeginLineValid)
92     return BeginLine;
93 
94   std::pair<FileID, unsigned> LocInfo = SM.getDecomposedLoc(Range.getBegin());
95   BeginLine = SM.getLineNumber(LocInfo.first, LocInfo.second);
96   BeginLineValid = true;
97   return BeginLine;
98 }
99 
100 unsigned RawComment::getEndLine(const SourceManager &SM) const {
101   if (EndLineValid)
102     return EndLine;
103 
104   std::pair<FileID, unsigned> LocInfo = SM.getDecomposedLoc(Range.getEnd());
105   EndLine = SM.getLineNumber(LocInfo.first, LocInfo.second);
106   EndLineValid = true;
107   return EndLine;
108 }
109 
110 StringRef RawComment::getRawTextSlow(const SourceManager &SourceMgr) const {
111   FileID BeginFileID;
112   FileID EndFileID;
113   unsigned BeginOffset;
114   unsigned EndOffset;
115 
116   llvm::tie(BeginFileID, BeginOffset) =
117       SourceMgr.getDecomposedLoc(Range.getBegin());
118   llvm::tie(EndFileID, EndOffset) =
119       SourceMgr.getDecomposedLoc(Range.getEnd());
120 
121   const unsigned Length = EndOffset - BeginOffset;
122   if (Length < 2)
123     return StringRef();
124 
125   // The comment can't begin in one file and end in another.
126   assert(BeginFileID == EndFileID);
127 
128   bool Invalid = false;
129   const char *BufferStart = SourceMgr.getBufferData(BeginFileID,
130                                                     &Invalid).data();
131   if (Invalid)
132     return StringRef();
133 
134   return StringRef(BufferStart + BeginOffset, Length);
135 }
136 
137 const char *RawComment::extractBriefText(const ASTContext &Context) const {
138   // Make sure that RawText is valid.
139   getRawText(Context.getSourceManager());
140 
141   // Since we will be copying the resulting text, all allocations made during
142   // parsing are garbage after resulting string is formed.  Thus we can use
143   // a separate allocator for all temporary stuff.
144   llvm::BumpPtrAllocator Allocator;
145 
146   comments::CommandTraits Traits;
147   comments::Lexer L(Allocator, Traits,
148                     Range.getBegin(), comments::CommentOptions(),
149                     RawText.begin(), RawText.end());
150   comments::BriefParser P(L, Traits);
151 
152   const std::string Result = P.Parse();
153   const unsigned BriefTextLength = Result.size();
154   char *BriefTextPtr = new (Context) char[BriefTextLength + 1];
155   memcpy(BriefTextPtr, Result.c_str(), BriefTextLength + 1);
156   BriefText = BriefTextPtr;
157   BriefTextValid = true;
158 
159   return BriefTextPtr;
160 }
161 
162 comments::FullComment *RawComment::parse(const ASTContext &Context,
163                                          const Decl *D) const {
164   // Make sure that RawText is valid.
165   getRawText(Context.getSourceManager());
166 
167   comments::CommandTraits Traits;
168   comments::Lexer L(Context.getAllocator(), Traits,
169                     getSourceRange().getBegin(), comments::CommentOptions(),
170                     RawText.begin(), RawText.end());
171   comments::Sema S(Context.getAllocator(), Context.getSourceManager(),
172                    Context.getDiagnostics(), Traits);
173   S.setDecl(D);
174   comments::Parser P(L, S, Context.getAllocator(), Context.getSourceManager(),
175                      Context.getDiagnostics(), Traits);
176 
177   return P.parseFullComment();
178 }
179 
180 namespace {
181 bool containsOnlyWhitespace(StringRef Str) {
182   return Str.find_first_not_of(" \t\f\v\r\n") == StringRef::npos;
183 }
184 
185 bool onlyWhitespaceBetweenComments(SourceManager &SM,
186                                    const RawComment &C1, const RawComment &C2) {
187   std::pair<FileID, unsigned> C1EndLocInfo = SM.getDecomposedLoc(
188                                                 C1.getSourceRange().getEnd());
189   std::pair<FileID, unsigned> C2BeginLocInfo = SM.getDecomposedLoc(
190                                               C2.getSourceRange().getBegin());
191 
192   // Question does not make sense if comments are located in different files.
193   if (C1EndLocInfo.first != C2BeginLocInfo.first)
194     return false;
195 
196   bool Invalid = false;
197   const char *Buffer = SM.getBufferData(C1EndLocInfo.first, &Invalid).data();
198   if (Invalid)
199     return false;
200 
201   StringRef TextBetweenComments(Buffer + C1EndLocInfo.second,
202                                 C2BeginLocInfo.second - C1EndLocInfo.second);
203 
204   return containsOnlyWhitespace(TextBetweenComments);
205 }
206 } // unnamed namespace
207 
208 void RawCommentList::addComment(const RawComment &RC,
209                                 llvm::BumpPtrAllocator &Allocator) {
210   if (RC.isInvalid())
211     return;
212 
213   // Check if the comments are not in source order.
214   while (!Comments.empty() &&
215          !SourceMgr.isBeforeInTranslationUnit(
216               Comments.back()->getSourceRange().getBegin(),
217               RC.getSourceRange().getBegin())) {
218     // If they are, just pop a few last comments that don't fit.
219     // This happens if an \#include directive contains comments.
220     Comments.pop_back();
221   }
222 
223   if (OnlyWhitespaceSeen) {
224     if (!onlyWhitespaceBetweenComments(SourceMgr, LastComment, RC))
225       OnlyWhitespaceSeen = false;
226   }
227 
228   LastComment = RC;
229 
230   // Ordinary comments are not interesting for us.
231   if (RC.isOrdinary())
232     return;
233 
234   // If this is the first Doxygen comment, save it (because there isn't
235   // anything to merge it with).
236   if (Comments.empty()) {
237     Comments.push_back(new (Allocator) RawComment(RC));
238     OnlyWhitespaceSeen = true;
239     return;
240   }
241 
242   const RawComment &C1 = *Comments.back();
243   const RawComment &C2 = RC;
244 
245   // Merge comments only if there is only whitespace between them.
246   // Can't merge trailing and non-trailing comments.
247   // Merge trailing comments if they are on same or consecutive lines.
248   if (OnlyWhitespaceSeen &&
249       (C1.isTrailingComment() == C2.isTrailingComment()) &&
250       (!C1.isTrailingComment() ||
251        C1.getEndLine(SourceMgr) + 1 >= C2.getBeginLine(SourceMgr))) {
252     SourceRange MergedRange(C1.getSourceRange().getBegin(),
253                             C2.getSourceRange().getEnd());
254     *Comments.back() = RawComment(SourceMgr, MergedRange, true);
255   } else
256     Comments.push_back(new (Allocator) RawComment(RC));
257 
258   OnlyWhitespaceSeen = true;
259 }
260 
261