1 //===--- IncludeCleaner.cpp - Unused/Missing Headers Analysis ---*- C++ -*-===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 
9 #include "IncludeCleaner.h"
10 #include "Config.h"
11 #include "Headers.h"
12 #include "ParsedAST.h"
13 #include "Protocol.h"
14 #include "SourceCode.h"
15 #include "support/Logger.h"
16 #include "support/Trace.h"
17 #include "clang/AST/ExprCXX.h"
18 #include "clang/AST/RecursiveASTVisitor.h"
19 #include "clang/Basic/SourceLocation.h"
20 #include "clang/Basic/SourceManager.h"
21 #include "clang/Lex/HeaderSearch.h"
22 #include "clang/Lex/Preprocessor.h"
23 #include "clang/Tooling/Syntax/Tokens.h"
24 #include "llvm/Support/FormatVariadic.h"
25 #include "llvm/Support/Path.h"
26 
27 namespace clang {
28 namespace clangd {
29 namespace {
30 
31 /// Crawler traverses the AST and feeds in the locations of (sometimes
32 /// implicitly) used symbols into \p Result.
33 class ReferencedLocationCrawler
34     : public RecursiveASTVisitor<ReferencedLocationCrawler> {
35 public:
36   ReferencedLocationCrawler(ReferencedLocations &Result,
37                             const SourceManager &SM)
38       : Result(Result), SM(SM) {}
39 
40   bool VisitDeclRefExpr(DeclRefExpr *DRE) {
41     add(DRE->getDecl());
42     add(DRE->getFoundDecl());
43     return true;
44   }
45 
46   bool VisitMemberExpr(MemberExpr *ME) {
47     add(ME->getMemberDecl());
48     add(ME->getFoundDecl().getDecl());
49     return true;
50   }
51 
52   bool VisitTagType(TagType *TT) {
53     add(TT->getDecl());
54     return true;
55   }
56 
57   bool VisitFunctionDecl(FunctionDecl *FD) {
58     // Function definition will require redeclarations to be included.
59     if (FD->isThisDeclarationADefinition())
60       add(FD);
61     return true;
62   }
63 
64   bool VisitCXXConstructExpr(CXXConstructExpr *CCE) {
65     add(CCE->getConstructor());
66     return true;
67   }
68 
69   bool VisitTemplateSpecializationType(TemplateSpecializationType *TST) {
70     if (isNew(TST)) {
71       add(TST->getTemplateName().getAsTemplateDecl()); // Primary template.
72       add(TST->getAsCXXRecordDecl());                  // Specialization
73     }
74     return true;
75   }
76 
77   bool VisitTypedefType(TypedefType *TT) {
78     add(TT->getDecl());
79     return true;
80   }
81 
82   // Consider types of any subexpression used, even if the type is not named.
83   // This is helpful in getFoo().bar(), where Foo must be complete.
84   // FIXME(kirillbobyrev): Should we tweak this? It may not be desirable to
85   // consider types "used" when they are not directly spelled in code.
86   bool VisitExpr(Expr *E) {
87     TraverseType(E->getType());
88     return true;
89   }
90 
91   bool TraverseType(QualType T) {
92     if (isNew(T.getTypePtrOrNull())) // don't care about quals
93       Base::TraverseType(T);
94     return true;
95   }
96 
97   bool VisitUsingDecl(UsingDecl *D) {
98     for (const auto *Shadow : D->shadows())
99       add(Shadow->getTargetDecl());
100     return true;
101   }
102 
103   // Enums may be usefully forward-declared as *complete* types by specifying
104   // an underlying type. In this case, the definition should see the declaration
105   // so they can be checked for compatibility.
106   bool VisitEnumDecl(EnumDecl *D) {
107     if (D->isThisDeclarationADefinition() && D->getIntegerTypeSourceInfo())
108       add(D);
109     return true;
110   }
111 
112   // When the overload is not resolved yet, mark all candidates as used.
113   bool VisitOverloadExpr(OverloadExpr *E) {
114     for (const auto *ResolutionDecl : E->decls())
115       add(ResolutionDecl);
116     return true;
117   }
118 
119 private:
120   using Base = RecursiveASTVisitor<ReferencedLocationCrawler>;
121 
122   void add(const Decl *D) {
123     if (!D || !isNew(D->getCanonicalDecl()))
124       return;
125     // Special case RecordDecls, as it is common for them to be forward
126     // declared multiple times. The most common cases are:
127     // - Definition available in TU, only mark that one as usage. The rest is
128     //   likely to be unnecessary. This might result in false positives when an
129     //   internal definition is visible.
130     // - There's a forward declaration in the main file, no need for other
131     //   redecls.
132     if (const auto *RD = llvm::dyn_cast<RecordDecl>(D)) {
133       if (const auto *Definition = RD->getDefinition()) {
134         Result.insert(Definition->getLocation());
135         return;
136       }
137       if (SM.isInMainFile(RD->getMostRecentDecl()->getLocation()))
138         return;
139     }
140     for (const Decl *Redecl : D->redecls())
141       Result.insert(Redecl->getLocation());
142   }
143 
144   bool isNew(const void *P) { return P && Visited.insert(P).second; }
145 
146   ReferencedLocations &Result;
147   llvm::DenseSet<const void *> Visited;
148   const SourceManager &SM;
149 };
150 
151 // Given a set of referenced FileIDs, determines all the potentially-referenced
152 // files and macros by traversing expansion/spelling locations of macro IDs.
153 // This is used to map the referenced SourceLocations onto real files.
154 struct ReferencedFiles {
155   ReferencedFiles(const SourceManager &SM) : SM(SM) {}
156   llvm::DenseSet<FileID> Files;
157   llvm::DenseSet<FileID> Macros;
158   const SourceManager &SM;
159 
160   void add(SourceLocation Loc) { add(SM.getFileID(Loc), Loc); }
161 
162   void add(FileID FID, SourceLocation Loc) {
163     if (FID.isInvalid())
164       return;
165     assert(SM.isInFileID(Loc, FID));
166     if (Loc.isFileID()) {
167       Files.insert(FID);
168       return;
169     }
170     // Don't process the same macro FID twice.
171     if (!Macros.insert(FID).second)
172       return;
173     const auto &Exp = SM.getSLocEntry(FID).getExpansion();
174     add(Exp.getSpellingLoc());
175     add(Exp.getExpansionLocStart());
176     add(Exp.getExpansionLocEnd());
177   }
178 };
179 
180 // Returns the range starting at '#' and ending at EOL. Escaped newlines are not
181 // handled.
182 clangd::Range getDiagnosticRange(llvm::StringRef Code, unsigned HashOffset) {
183   clangd::Range Result;
184   Result.end = Result.start = offsetToPosition(Code, HashOffset);
185 
186   // Span the warning until the EOL or EOF.
187   Result.end.character +=
188       lspLength(Code.drop_front(HashOffset).take_until([](char C) {
189         return C == '\n' || C == '\r';
190       }));
191   return Result;
192 }
193 
194 // Finds locations of macros referenced from within the main file. That includes
195 // references that were not yet expanded, e.g `BAR` in `#define FOO BAR`.
196 void findReferencedMacros(ParsedAST &AST, ReferencedLocations &Result) {
197   trace::Span Tracer("IncludeCleaner::findReferencedMacros");
198   auto &SM = AST.getSourceManager();
199   auto &PP = AST.getPreprocessor();
200   // FIXME(kirillbobyrev): The macros from the main file are collected in
201   // ParsedAST's MainFileMacros. However, we can't use it here because it
202   // doesn't handle macro references that were not expanded, e.g. in macro
203   // definitions or preprocessor-disabled sections.
204   //
205   // Extending MainFileMacros to collect missing references and switching to
206   // this mechanism (as opposed to iterating through all tokens) will improve
207   // the performance of findReferencedMacros and also improve other features
208   // relying on MainFileMacros.
209   for (const syntax::Token &Tok :
210        AST.getTokens().spelledTokens(SM.getMainFileID())) {
211     auto Macro = locateMacroAt(Tok, PP);
212     if (!Macro)
213       continue;
214     auto Loc = Macro->Info->getDefinitionLoc();
215     if (Loc.isValid())
216       Result.insert(Loc);
217   }
218 }
219 
220 bool mayConsiderUnused(const Inclusion &Inc, ParsedAST &AST) {
221   // FIXME(kirillbobyrev): We currently do not support the umbrella headers.
222   // Standard Library headers are typically umbrella headers, and system
223   // headers are likely to be the Standard Library headers. Until we have a
224   // good support for umbrella headers and Standard Library headers, don't warn
225   // about them.
226   if (Inc.Written.front() == '<' || Inc.BehindPragmaKeep)
227     return false;
228   // Headers without include guards have side effects and are not
229   // self-contained, skip them.
230   assert(Inc.HeaderID);
231   auto FE = AST.getSourceManager().getFileManager().getFile(
232       AST.getIncludeStructure().getRealPath(
233           static_cast<IncludeStructure::HeaderID>(*Inc.HeaderID)));
234   assert(FE);
235   if (!AST.getPreprocessor().getHeaderSearchInfo().isFileMultipleIncludeGuarded(
236           *FE)) {
237     dlog("{0} doesn't have header guard and will not be considered unused",
238          (*FE)->getName());
239     return false;
240   }
241   return true;
242 }
243 
244 // In case symbols are coming from non self-contained header, we need to find
245 // its first includer that is self-contained. This is the header users can
246 // include, so it will be responsible for bringing the symbols from given
247 // header into the scope.
248 FileID headerResponsible(FileID ID, const SourceManager &SM,
249                          const IncludeStructure &Includes) {
250   // Unroll the chain of non self-contained headers until we find the one that
251   // can be included.
252   for (const FileEntry *FE = SM.getFileEntryForID(ID); ID != SM.getMainFileID();
253        FE = SM.getFileEntryForID(ID)) {
254     // If FE is nullptr, we consider it to be the responsible header.
255     if (!FE)
256       break;
257     auto HID = Includes.getID(FE);
258     assert(HID && "We're iterating over headers already existing in "
259                   "IncludeStructure");
260     if (Includes.isSelfContained(*HID))
261       break;
262     // The header is not self-contained: put the responsibility for its symbols
263     // on its includer.
264     ID = SM.getFileID(SM.getIncludeLoc(ID));
265   }
266   return ID;
267 }
268 
269 } // namespace
270 
271 ReferencedLocations findReferencedLocations(ParsedAST &AST) {
272   trace::Span Tracer("IncludeCleaner::findReferencedLocations");
273   ReferencedLocations Result;
274   ReferencedLocationCrawler Crawler(Result, AST.getSourceManager());
275   Crawler.TraverseAST(AST.getASTContext());
276   findReferencedMacros(AST, Result);
277   return Result;
278 }
279 
280 llvm::DenseSet<FileID>
281 findReferencedFiles(const llvm::DenseSet<SourceLocation> &Locs,
282                     const IncludeStructure &Includes, const SourceManager &SM) {
283   std::vector<SourceLocation> Sorted{Locs.begin(), Locs.end()};
284   llvm::sort(Sorted); // Group by FileID.
285   ReferencedFiles Files(SM);
286   for (auto It = Sorted.begin(); It < Sorted.end();) {
287     FileID FID = SM.getFileID(*It);
288     Files.add(FID, *It);
289     // Cheaply skip over all the other locations from the same FileID.
290     // This avoids lots of redundant Loc->File lookups for the same file.
291     do
292       ++It;
293     while (It != Sorted.end() && SM.isInFileID(*It, FID));
294   }
295   // If a header is not self-contained, we consider its symbols a logical part
296   // of the including file. Therefore, mark the parents of all used
297   // non-self-contained FileIDs as used. Perform this on FileIDs rather than
298   // HeaderIDs, as each inclusion of a non-self-contained file is distinct.
299   llvm::DenseSet<FileID> Result;
300   for (FileID ID : Files.Files)
301     Result.insert(headerResponsible(ID, SM, Includes));
302   return Result;
303 }
304 
305 std::vector<const Inclusion *>
306 getUnused(ParsedAST &AST,
307           const llvm::DenseSet<IncludeStructure::HeaderID> &ReferencedFiles) {
308   trace::Span Tracer("IncludeCleaner::getUnused");
309   std::vector<const Inclusion *> Unused;
310   for (const Inclusion &MFI : AST.getIncludeStructure().MainFileIncludes) {
311     if (!MFI.HeaderID)
312       continue;
313     auto IncludeID = static_cast<IncludeStructure::HeaderID>(*MFI.HeaderID);
314     bool Used = ReferencedFiles.contains(IncludeID);
315     if (!Used && !mayConsiderUnused(MFI, AST)) {
316       dlog("{0} was not used, but is not eligible to be diagnosed as unused",
317            MFI.Written);
318       continue;
319     }
320     if (!Used)
321       Unused.push_back(&MFI);
322     dlog("{0} is {1}", MFI.Written, Used ? "USED" : "UNUSED");
323   }
324   return Unused;
325 }
326 
327 #ifndef NDEBUG
328 // Is FID a <built-in>, <scratch space> etc?
329 static bool isSpecialBuffer(FileID FID, const SourceManager &SM) {
330   const SrcMgr::FileInfo &FI = SM.getSLocEntry(FID).getFile();
331   return FI.getName().startswith("<");
332 }
333 #endif
334 
335 llvm::DenseSet<IncludeStructure::HeaderID>
336 translateToHeaderIDs(const llvm::DenseSet<FileID> &Files,
337                      const IncludeStructure &Includes,
338                      const SourceManager &SM) {
339   trace::Span Tracer("IncludeCleaner::translateToHeaderIDs");
340   llvm::DenseSet<IncludeStructure::HeaderID> TranslatedHeaderIDs;
341   TranslatedHeaderIDs.reserve(Files.size());
342   for (FileID FID : Files) {
343     const FileEntry *FE = SM.getFileEntryForID(FID);
344     if (!FE) {
345       assert(isSpecialBuffer(FID, SM));
346       continue;
347     }
348     const auto File = Includes.getID(FE);
349     assert(File);
350     TranslatedHeaderIDs.insert(*File);
351   }
352   return TranslatedHeaderIDs;
353 }
354 
355 std::vector<const Inclusion *> computeUnusedIncludes(ParsedAST &AST) {
356   const auto &SM = AST.getSourceManager();
357 
358   auto Refs = findReferencedLocations(AST);
359   auto ReferencedFileIDs = findReferencedFiles(Refs, AST.getIncludeStructure(),
360                                                AST.getSourceManager());
361   auto ReferencedHeaders =
362       translateToHeaderIDs(ReferencedFileIDs, AST.getIncludeStructure(), SM);
363   return getUnused(AST, ReferencedHeaders);
364 }
365 
366 std::vector<Diag> issueUnusedIncludesDiagnostics(ParsedAST &AST,
367                                                  llvm::StringRef Code) {
368   const Config &Cfg = Config::current();
369   if (Cfg.Diagnostics.UnusedIncludes != Config::UnusedIncludesPolicy::Strict ||
370       Cfg.Diagnostics.SuppressAll ||
371       Cfg.Diagnostics.Suppress.contains("unused-includes"))
372     return {};
373   trace::Span Tracer("IncludeCleaner::issueUnusedIncludesDiagnostics");
374   std::vector<Diag> Result;
375   std::string FileName =
376       AST.getSourceManager()
377           .getFileEntryForID(AST.getSourceManager().getMainFileID())
378           ->getName()
379           .str();
380   for (const auto *Inc : computeUnusedIncludes(AST)) {
381     Diag D;
382     D.Message =
383         llvm::formatv("included header {0} is not used",
384                       llvm::sys::path::filename(
385                           Inc->Written.substr(1, Inc->Written.size() - 2),
386                           llvm::sys::path::Style::posix));
387     D.Name = "unused-includes";
388     D.Source = Diag::DiagSource::Clangd;
389     D.File = FileName;
390     D.Severity = DiagnosticsEngine::Warning;
391     D.Tags.push_back(Unnecessary);
392     D.Range = getDiagnosticRange(Code, Inc->HashOffset);
393     // FIXME(kirillbobyrev): Removing inclusion might break the code if the
394     // used headers are only reachable transitively through this one. Suggest
395     // including them directly instead.
396     // FIXME(kirillbobyrev): Add fix suggestion for adding IWYU pragmas
397     // (keep/export) remove the warning once we support IWYU pragmas.
398     D.Fixes.emplace_back();
399     D.Fixes.back().Message = "remove #include directive";
400     D.Fixes.back().Edits.emplace_back();
401     D.Fixes.back().Edits.back().range.start.line = Inc->HashLine;
402     D.Fixes.back().Edits.back().range.end.line = Inc->HashLine + 1;
403     D.InsideMainFile = true;
404     Result.push_back(std::move(D));
405   }
406   return Result;
407 }
408 
409 } // namespace clangd
410 } // namespace clang
411