1 //===-- SymbolIndexManager.cpp - Managing multiple SymbolIndices-*- 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 "SymbolIndexManager.h" 10 #include "find-all-symbols/SymbolInfo.h" 11 #include "llvm/ADT/STLExtras.h" 12 #include "llvm/ADT/SmallVector.h" 13 #include "llvm/ADT/StringMap.h" 14 #include "llvm/Support/Debug.h" 15 #include "llvm/Support/Path.h" 16 17 #define DEBUG_TYPE "clang-include-fixer" 18 19 namespace clang { 20 namespace include_fixer { 21 22 using find_all_symbols::SymbolInfo; 23 using find_all_symbols::SymbolAndSignals; 24 25 // Calculate a score based on whether we think the given header is closely 26 // related to the given source file. 27 static double similarityScore(llvm::StringRef FileName, 28 llvm::StringRef Header) { 29 // Compute the maximum number of common path segments between Header and 30 // a suffix of FileName. 31 // We do not do a full longest common substring computation, as Header 32 // specifies the path we would directly #include, so we assume it is rooted 33 // relatively to a subproject of the repository. 34 int MaxSegments = 1; 35 for (auto FileI = llvm::sys::path::begin(FileName), 36 FileE = llvm::sys::path::end(FileName); 37 FileI != FileE; ++FileI) { 38 int Segments = 0; 39 for (auto HeaderI = llvm::sys::path::begin(Header), 40 HeaderE = llvm::sys::path::end(Header), I = FileI; 41 HeaderI != HeaderE && *I == *HeaderI && I != FileE; ++I, ++HeaderI) { 42 ++Segments; 43 } 44 MaxSegments = std::max(Segments, MaxSegments); 45 } 46 return MaxSegments; 47 } 48 49 static void rank(std::vector<SymbolAndSignals> &Symbols, 50 llvm::StringRef FileName) { 51 llvm::StringMap<double> Score; 52 for (const auto &Symbol : Symbols) { 53 // Calculate a score from the similarity of the header the symbol is in 54 // with the current file and the popularity of the symbol. 55 double NewScore = similarityScore(FileName, Symbol.Symbol.getFilePath()) * 56 (1.0 + std::log2(1 + Symbol.Signals.Seen)); 57 double &S = Score[Symbol.Symbol.getFilePath()]; 58 S = std::max(S, NewScore); 59 } 60 // Sort by the gathered scores. Use file name as a tie breaker so we can 61 // deduplicate. 62 llvm::sort(Symbols.begin(), Symbols.end(), 63 [&](const SymbolAndSignals &A, const SymbolAndSignals &B) { 64 auto AS = Score[A.Symbol.getFilePath()]; 65 auto BS = Score[B.Symbol.getFilePath()]; 66 if (AS != BS) 67 return AS > BS; 68 return A.Symbol.getFilePath() < B.Symbol.getFilePath(); 69 }); 70 } 71 72 std::vector<find_all_symbols::SymbolInfo> 73 SymbolIndexManager::search(llvm::StringRef Identifier, 74 bool IsNestedSearch, 75 llvm::StringRef FileName) const { 76 // The identifier may be fully qualified, so split it and get all the context 77 // names. 78 llvm::SmallVector<llvm::StringRef, 8> Names; 79 Identifier.split(Names, "::"); 80 81 bool IsFullyQualified = false; 82 if (Identifier.startswith("::")) { 83 Names.erase(Names.begin()); // Drop first (empty) element. 84 IsFullyQualified = true; 85 } 86 87 // As long as we don't find a result keep stripping name parts from the end. 88 // This is to support nested classes which aren't recorded in the database. 89 // Eventually we will either hit a class (namespaces aren't in the database 90 // either) and can report that result. 91 bool TookPrefix = false; 92 std::vector<SymbolAndSignals> MatchedSymbols; 93 do { 94 std::vector<SymbolAndSignals> Symbols; 95 for (const auto &DB : SymbolIndices) { 96 auto Res = DB.get()->search(Names.back()); 97 Symbols.insert(Symbols.end(), Res.begin(), Res.end()); 98 } 99 100 LLVM_DEBUG(llvm::dbgs() << "Searching " << Names.back() << "... got " 101 << Symbols.size() << " results...\n"); 102 103 for (auto &SymAndSig : Symbols) { 104 const SymbolInfo &Symbol = SymAndSig.Symbol; 105 // Match the identifier name without qualifier. 106 bool IsMatched = true; 107 auto SymbolContext = Symbol.getContexts().begin(); 108 auto IdentiferContext = Names.rbegin() + 1; // Skip identifier name. 109 // Match the remaining context names. 110 while (IdentiferContext != Names.rend() && 111 SymbolContext != Symbol.getContexts().end()) { 112 if (SymbolContext->second == *IdentiferContext) { 113 ++IdentiferContext; 114 ++SymbolContext; 115 } else if (SymbolContext->first == 116 find_all_symbols::SymbolInfo::ContextType::EnumDecl) { 117 // Skip non-scoped enum context. 118 ++SymbolContext; 119 } else { 120 IsMatched = false; 121 break; 122 } 123 } 124 125 // If the name was qualified we only want to add results if we evaluated 126 // all contexts. 127 if (IsFullyQualified) 128 IsMatched &= (SymbolContext == Symbol.getContexts().end()); 129 130 // FIXME: Support full match. At this point, we only find symbols in 131 // database which end with the same contexts with the identifier. 132 if (IsMatched && IdentiferContext == Names.rend()) { 133 // If we're in a situation where we took a prefix but the thing we 134 // found couldn't possibly have a nested member ignore it. 135 if (TookPrefix && 136 (Symbol.getSymbolKind() == SymbolInfo::SymbolKind::Function || 137 Symbol.getSymbolKind() == SymbolInfo::SymbolKind::Variable || 138 Symbol.getSymbolKind() == 139 SymbolInfo::SymbolKind::EnumConstantDecl || 140 Symbol.getSymbolKind() == SymbolInfo::SymbolKind::Macro)) 141 continue; 142 143 MatchedSymbols.push_back(std::move(SymAndSig)); 144 } 145 } 146 Names.pop_back(); 147 TookPrefix = true; 148 } while (MatchedSymbols.empty() && !Names.empty() && IsNestedSearch); 149 150 rank(MatchedSymbols, FileName); 151 // Strip signals, they are no longer needed. 152 std::vector<SymbolInfo> Res; 153 Res.reserve(MatchedSymbols.size()); 154 for (auto &SymAndSig : MatchedSymbols) 155 Res.push_back(std::move(SymAndSig.Symbol)); 156 return Res; 157 } 158 159 } // namespace include_fixer 160 } // namespace clang 161