| 1 | //===--- FileDistance.cpp - File contents container -------------*- 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 | // The FileDistance structure allows calculating the minimum distance to paths |
| 10 | // in a single tree. |
| 11 | // We simply walk up the path's ancestors until we find a node whose cost is |
| 12 | // known, and add the cost of walking back down. Initialization ensures this |
| 13 | // gives the correct path to the roots. |
| 14 | // We cache the results, so that the runtime is O(|A|), where A is the set of |
| 15 | // all distinct ancestors of visited paths. |
| 16 | // |
| 17 | // Example after initialization with /=2, /bar=0, DownCost = 1: |
| 18 | // / = 2 |
| 19 | // /bar = 0 |
| 20 | // |
| 21 | // After querying /foo/bar and /bar/foo: |
| 22 | // / = 2 |
| 23 | // /bar = 0 |
| 24 | // /bar/foo = 1 |
| 25 | // /foo = 3 |
| 26 | // /foo/bar = 4 |
| 27 | // |
| 28 | // URIDistance creates FileDistance lazily for each URI scheme encountered. In |
| 29 | // practice this is a small constant factor. |
| 30 | // |
| 31 | //===-------------------------------------------------------------------------// |
| 32 | |
| 33 | #include "FileDistance.h" |
| 34 | #include "URI.h" |
| 35 | #include "support/Logger.h" |
| 36 | #include "llvm/ADT/STLExtras.h" |
| 37 | #include "llvm/ADT/StringExtras.h" |
| 38 | #include "llvm/Support/Path.h" |
| 39 | #include <queue> |
| 40 | |
| 41 | namespace clang { |
| 42 | namespace clangd { |
| 43 | |
| 44 | // Convert a path into the canonical form. |
| 45 | // Canonical form is either "/", or "/segment" * N: |
| 46 | // C:\foo\bar --> /c:/foo/bar |
| 47 | // /foo/ --> /foo |
| 48 | // a/b/c --> /a/b/c |
| 49 | static llvm::SmallString<128> canonicalize(llvm::StringRef Path) { |
| 50 | llvm::SmallString<128> Result = Path.rtrim(Char: '/'); |
| 51 | native(path&: Result, style: llvm::sys::path::Style::posix); |
| 52 | if (Result.empty() || Result.front() != '/') |
| 53 | Result.insert(I: Result.begin(), Elt: '/'); |
| 54 | return Result; |
| 55 | } |
| 56 | |
| 57 | constexpr const unsigned FileDistance::Unreachable; |
| 58 | const llvm::hash_code FileDistance::RootHash = |
| 59 | llvm::hash_value(S: llvm::StringRef("/" )); |
| 60 | |
| 61 | FileDistance::FileDistance(llvm::StringMap<SourceParams> Sources, |
| 62 | const FileDistanceOptions &Opts) |
| 63 | : Opts(Opts) { |
| 64 | llvm::DenseMap<llvm::hash_code, llvm::SmallVector<llvm::hash_code>> DownEdges; |
| 65 | // Compute the best distance following only up edges. |
| 66 | // Keep track of down edges, in case we can use them to improve on this. |
| 67 | for (const auto &S : Sources) { |
| 68 | auto Canonical = canonicalize(Path: S.getKey()); |
| 69 | dlog("Source {0} = {1}, MaxUp = {2}" , Canonical, S.second.Cost, |
| 70 | S.second.MaxUpTraversals); |
| 71 | // Walk up to ancestors of this source, assigning cost. |
| 72 | llvm::StringRef Rest = Canonical; |
| 73 | llvm::hash_code Hash = llvm::hash_value(S: Rest); |
| 74 | for (unsigned I = 0; !Rest.empty(); ++I) { |
| 75 | Rest = parent_path(path: Rest, style: llvm::sys::path::Style::posix); |
| 76 | auto NextHash = llvm::hash_value(S: Rest); |
| 77 | auto &Down = DownEdges[NextHash]; |
| 78 | if (!llvm::is_contained(Range&: Down, Element: Hash)) |
| 79 | Down.push_back(Elt: Hash); |
| 80 | // We can't just break after MaxUpTraversals, must still set DownEdges. |
| 81 | if (I > S.getValue().MaxUpTraversals) { |
| 82 | if (Cache.contains(Val: Hash)) |
| 83 | break; |
| 84 | } else { |
| 85 | unsigned Cost = S.getValue().Cost + I * Opts.UpCost; |
| 86 | auto R = Cache.try_emplace(Key: Hash, Args&: Cost); |
| 87 | if (!R.second) { |
| 88 | if (Cost < R.first->second) { |
| 89 | R.first->second = Cost; |
| 90 | } else { |
| 91 | // If we're not the best way to get to this path, stop assigning. |
| 92 | break; |
| 93 | } |
| 94 | } |
| 95 | } |
| 96 | Hash = NextHash; |
| 97 | } |
| 98 | } |
| 99 | // Now propagate scores parent -> child if that's an improvement. |
| 100 | // BFS ensures we propagate down chains (must visit parents before children). |
| 101 | std::queue<llvm::hash_code> Next; |
| 102 | for (auto Child : DownEdges.lookup(Val: llvm::hash_value(S: llvm::StringRef("" )))) |
| 103 | Next.push(x: Child); |
| 104 | while (!Next.empty()) { |
| 105 | auto Parent = Next.front(); |
| 106 | Next.pop(); |
| 107 | auto ParentCost = Cache.lookup(Val: Parent); |
| 108 | for (auto Child : DownEdges.lookup(Val: Parent)) { |
| 109 | if (Parent != RootHash || Opts.AllowDownTraversalFromRoot) { |
| 110 | auto &ChildCost = |
| 111 | Cache.try_emplace(Key: Child, Args: Unreachable).first->getSecond(); |
| 112 | if (ParentCost + Opts.DownCost < ChildCost) |
| 113 | ChildCost = ParentCost + Opts.DownCost; |
| 114 | } |
| 115 | Next.push(x: Child); |
| 116 | } |
| 117 | } |
| 118 | } |
| 119 | |
| 120 | unsigned FileDistance::distance(llvm::StringRef Path) { |
| 121 | auto Canonical = canonicalize(Path); |
| 122 | unsigned Cost = Unreachable; |
| 123 | llvm::SmallVector<llvm::hash_code> Ancestors; |
| 124 | // Walk up ancestors until we find a path we know the distance for. |
| 125 | for (llvm::StringRef Rest = Canonical; !Rest.empty(); |
| 126 | Rest = parent_path(path: Rest, style: llvm::sys::path::Style::posix)) { |
| 127 | auto Hash = llvm::hash_value(S: Rest); |
| 128 | if (Hash == RootHash && !Ancestors.empty() && |
| 129 | !Opts.AllowDownTraversalFromRoot) { |
| 130 | Cost = Unreachable; |
| 131 | break; |
| 132 | } |
| 133 | auto It = Cache.find(Val: Hash); |
| 134 | if (It != Cache.end()) { |
| 135 | Cost = It->second; |
| 136 | break; |
| 137 | } |
| 138 | Ancestors.push_back(Elt: Hash); |
| 139 | } |
| 140 | // Now we know the costs for (known node, queried node]. |
| 141 | // Fill these in, walking down the directory tree. |
| 142 | for (llvm::hash_code Hash : llvm::reverse(C&: Ancestors)) { |
| 143 | if (Cost != Unreachable) |
| 144 | Cost += Opts.DownCost; |
| 145 | Cache.try_emplace(Key: Hash, Args&: Cost); |
| 146 | } |
| 147 | dlog("distance({0} = {1})" , Path, Cost); |
| 148 | return Cost; |
| 149 | } |
| 150 | |
| 151 | unsigned URIDistance::distance(llvm::StringRef URI) { |
| 152 | auto R = Cache.try_emplace(Key: llvm::hash_value(S: URI), Args: FileDistance::Unreachable); |
| 153 | if (!R.second) |
| 154 | return R.first->getSecond(); |
| 155 | if (auto U = clangd::URI::parse(Uri: URI)) { |
| 156 | dlog("distance({0} = {1})" , URI, U->body()); |
| 157 | R.first->second = forScheme(Scheme: U->scheme()).distance(Path: U->body()); |
| 158 | } else { |
| 159 | log(Fmt: "URIDistance::distance() of unparseable {0}: {1}" , Vals&: URI, Vals: U.takeError()); |
| 160 | } |
| 161 | return R.first->second; |
| 162 | } |
| 163 | |
| 164 | FileDistance &URIDistance::forScheme(llvm::StringRef Scheme) { |
| 165 | auto &Delegate = ByScheme[Scheme]; |
| 166 | if (!Delegate) { |
| 167 | llvm::StringMap<SourceParams> SchemeSources; |
| 168 | for (const auto &Source : Sources) { |
| 169 | if (auto U = clangd::URI::create(AbsolutePath: Source.getKey(), Scheme)) |
| 170 | SchemeSources.try_emplace(Key: U->body(), Args: Source.getValue()); |
| 171 | else |
| 172 | llvm::consumeError(Err: U.takeError()); |
| 173 | } |
| 174 | dlog("FileDistance for scheme {0}: {1}/{2} sources" , Scheme, |
| 175 | SchemeSources.size(), Sources.size()); |
| 176 | Delegate.reset(p: new FileDistance(std::move(SchemeSources), Opts)); |
| 177 | } |
| 178 | return *Delegate; |
| 179 | } |
| 180 | |
| 181 | static std::pair<std::string, int> scopeToPath(llvm::StringRef Scope) { |
| 182 | llvm::SmallVector<llvm::StringRef> Split; |
| 183 | Scope.split(A&: Split, Separator: "::" , /*MaxSplit=*/-1, /*KeepEmpty=*/false); |
| 184 | return {"/" + llvm::join(R&: Split, Separator: "/" ), Split.size()}; |
| 185 | } |
| 186 | |
| 187 | static FileDistance |
| 188 | createScopeFileDistance(llvm::ArrayRef<std::string> QueryScopes) { |
| 189 | FileDistanceOptions Opts; |
| 190 | Opts.UpCost = 2; |
| 191 | Opts.DownCost = 4; |
| 192 | Opts.AllowDownTraversalFromRoot = false; |
| 193 | |
| 194 | llvm::StringMap<SourceParams> Sources; |
| 195 | llvm::StringRef Preferred = |
| 196 | QueryScopes.empty() ? "" : QueryScopes.front().c_str(); |
| 197 | for (llvm::StringRef S : QueryScopes) { |
| 198 | SourceParams Param; |
| 199 | // Penalize the global scope even it's preferred, as all projects can define |
| 200 | // symbols in it, and there is pattern where using-namespace is used in |
| 201 | // place of enclosing namespaces (e.g. in implementation files). |
| 202 | if (S == Preferred) |
| 203 | Param.Cost = S == "" ? 4 : 0; |
| 204 | else if (Preferred.starts_with(Prefix: S) && !S.empty()) |
| 205 | continue; // just rely on up-traversals. |
| 206 | else |
| 207 | Param.Cost = S == "" ? 6 : 2; |
| 208 | auto Path = scopeToPath(Scope: S); |
| 209 | // The global namespace is not 'near' its children. |
| 210 | Param.MaxUpTraversals = std::max(a: Path.second - 1, b: 0); |
| 211 | Sources[Path.first] = std::move(Param); |
| 212 | } |
| 213 | return FileDistance(std::move(Sources), Opts); |
| 214 | } |
| 215 | |
| 216 | ScopeDistance::ScopeDistance(llvm::ArrayRef<std::string> QueryScopes) |
| 217 | : Distance(createScopeFileDistance(QueryScopes)) {} |
| 218 | |
| 219 | unsigned ScopeDistance::distance(llvm::StringRef SymbolScope) { |
| 220 | return Distance.distance(Path: scopeToPath(Scope: SymbolScope).first); |
| 221 | } |
| 222 | |
| 223 | } // namespace clangd |
| 224 | } // namespace clang |
| 225 | |