| 1 | //===--- Selection.h - What's under the cursor? -------------------*-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 | // Many features are triggered at locations/ranges and operate on AST nodes. |
| 9 | // (e.g. go-to-definition or code tweaks). |
| 10 | // At a high level, such features need to work out which node is the correct |
| 11 | // target. |
| 12 | // |
| 13 | // There are a few levels of ambiguity here: |
| 14 | // |
| 15 | // Which tokens are included: |
| 16 | // int x = one + two; // what should "go to definition" do? |
| 17 | // ^^^^^^ |
| 18 | // |
| 19 | // Same token means multiple things: |
| 20 | // string("foo") // class string, or a constructor? |
| 21 | // ^ |
| 22 | // |
| 23 | // Which level of the AST is interesting? |
| 24 | // if (err) { // reference to 'err', or operator bool(), |
| 25 | // ^ // or the if statement itself? |
| 26 | // |
| 27 | // Here we build and expose a data structure that allows features to resolve |
| 28 | // these ambiguities in an appropriate way: |
| 29 | // - we determine which low-level nodes are partly or completely covered |
| 30 | // by the selection. |
| 31 | // - we expose a tree of the selected nodes and their lexical parents. |
| 32 | // |
| 33 | // Sadly LSP specifies locations as being between characters, and this causes |
| 34 | // some ambiguities we cannot cleanly resolve: |
| 35 | // lhs+rhs // targeting '+' or 'lhs'? |
| 36 | // ^ // in GUI editors, double-clicking 'lhs' yields this position! |
| 37 | // |
| 38 | // The best we can do in these cases is try both, which leads to the awkward |
| 39 | // SelectionTree::createEach() API. |
| 40 | //===----------------------------------------------------------------------===// |
| 41 | |
| 42 | #ifndef LLVM_CLANG_TOOLS_EXTRA_CLANGD_SELECTION_H |
| 43 | #define |
| 44 | #include "clang/AST/ASTTypeTraits.h" |
| 45 | #include "clang/AST/PrettyPrinter.h" |
| 46 | #include "clang/Tooling/Syntax/Tokens.h" |
| 47 | #include "llvm/ADT/SmallVector.h" |
| 48 | #include <stack> |
| 49 | |
| 50 | namespace clang { |
| 51 | namespace clangd { |
| 52 | |
| 53 | // A selection can partially or completely cover several AST nodes. |
| 54 | // The SelectionTree contains nodes that are covered, and their parents. |
| 55 | // SelectionTree does not contain all AST nodes, rather only: |
| 56 | // Decl, Stmt, TypeLoc, NestedNamespaceSpecifierLoc, CXXCtorInitializer. |
| 57 | // (These are the nodes with source ranges that fit in DynTypedNode). |
| 58 | // |
| 59 | // Usually commonAncestor() is the place to start: |
| 60 | // - it's the simplest answer to "what node is under the cursor" |
| 61 | // - the selected Expr (for example) can be found by walking up the parent |
| 62 | // chain and checking Node->ASTNode. |
| 63 | // - if you want to traverse the selected nodes, they are all under |
| 64 | // commonAncestor() in the tree. |
| 65 | // |
| 66 | // SelectionTree tries to behave sensibly in the presence of macros, but does |
| 67 | // not model any preprocessor concepts: the output is a subset of the AST. |
| 68 | // When a macro argument is specifically selected, only its first expansion is |
| 69 | // selected in the AST. (Returning a selection forest is unreasonably difficult |
| 70 | // for callers to handle correctly.) |
| 71 | // |
| 72 | // Comments, directives and whitespace are completely ignored. |
| 73 | // Semicolons are also ignored, as the AST generally does not model them well. |
| 74 | // |
| 75 | // The SelectionTree owns the Node structures, but the ASTNode attributes |
| 76 | // point back into the AST it was constructed with. |
| 77 | class SelectionTree { |
| 78 | public: |
| 79 | // Create selection trees for the given range, and pass them to Func. |
| 80 | // |
| 81 | // There may be multiple possible selection trees: |
| 82 | // - if the range is empty and borders two tokens, a tree for the right token |
| 83 | // and a tree for the left token will be yielded. |
| 84 | // - Func should return true on success (stop) and false on failure (continue) |
| 85 | // |
| 86 | // Always yields at least one tree. If no tokens are touched, it is empty. |
| 87 | static bool createEach(ASTContext &AST, const syntax::TokenBuffer &Tokens, |
| 88 | unsigned Begin, unsigned End, |
| 89 | llvm::function_ref<bool(SelectionTree)> Func); |
| 90 | |
| 91 | // Create a selection tree for the given range. |
| 92 | // |
| 93 | // Where ambiguous (range is empty and borders two tokens), prefer the token |
| 94 | // on the right. |
| 95 | static SelectionTree createRight(ASTContext &AST, |
| 96 | const syntax::TokenBuffer &Tokens, |
| 97 | unsigned Begin, unsigned End); |
| 98 | |
| 99 | // Copies are no good - contain pointers to other nodes. |
| 100 | SelectionTree(const SelectionTree &) = delete; |
| 101 | SelectionTree &operator=(const SelectionTree &) = delete; |
| 102 | // Moves are OK though - internal storage is pointer-stable when moved. |
| 103 | SelectionTree(SelectionTree &&) = default; |
| 104 | SelectionTree &operator=(SelectionTree &&) = default; |
| 105 | |
| 106 | // Describes to what extent an AST node is covered by the selection. |
| 107 | enum Selection : unsigned char { |
| 108 | // The AST node owns no characters covered by the selection. |
| 109 | // Note that characters owned by children don't count: |
| 110 | // if (x == 0) scream(); |
| 111 | // ^^^^^^ |
| 112 | // The IfStmt would be Unselected because all the selected characters are |
| 113 | // associated with its children. |
| 114 | // (Invisible nodes like ImplicitCastExpr are always unselected). |
| 115 | Unselected, |
| 116 | // The AST node owns selected characters, but is not completely covered. |
| 117 | Partial, |
| 118 | // The AST node owns characters, and is covered by the selection. |
| 119 | Complete, |
| 120 | }; |
| 121 | // An AST node that is implicated in the selection. |
| 122 | // (Either selected directly, or some descendant is selected). |
| 123 | struct Node { |
| 124 | // The parent within the selection tree. nullptr for TranslationUnitDecl. |
| 125 | Node *Parent; |
| 126 | // Direct children within the selection tree. |
| 127 | llvm::SmallVector<const Node *> Children; |
| 128 | // The corresponding node from the full AST. |
| 129 | DynTypedNode ASTNode; |
| 130 | // The extent to which this node is covered by the selection. |
| 131 | Selection Selected; |
| 132 | // Walk up the AST to get the lexical DeclContext of this Node, which is not |
| 133 | // the node itself. |
| 134 | const DeclContext &getDeclContext() const; |
| 135 | // Printable node kind, like "CXXRecordDecl" or "AutoTypeLoc". |
| 136 | std::string kind() const; |
| 137 | // If this node is a wrapper with no syntax (e.g. implicit cast), return |
| 138 | // its contents. (If multiple wrappers are present, unwraps all of them). |
| 139 | const Node &ignoreImplicit() const; |
| 140 | // If this node is inside a wrapper with no syntax (e.g. implicit cast), |
| 141 | // return that wrapper. (If multiple are present, unwraps all of them). |
| 142 | const Node &outerImplicit() const; |
| 143 | }; |
| 144 | // The most specific common ancestor of all the selected nodes. |
| 145 | // Returns nullptr if the common ancestor is the root. |
| 146 | // (This is to avoid accidentally traversing the TUDecl and thus preamble). |
| 147 | const Node *commonAncestor() const; |
| 148 | // The selection node corresponding to TranslationUnitDecl. |
| 149 | const Node &root() const { return *Root; } |
| 150 | |
| 151 | private: |
| 152 | // Creates a selection tree for the given range in the main file. |
| 153 | // The range includes bytes [Start, End). |
| 154 | SelectionTree(ASTContext &AST, const syntax::TokenBuffer &Tokens, |
| 155 | unsigned Start, unsigned End); |
| 156 | |
| 157 | std::deque<Node> Nodes; // Stable-pointer storage. |
| 158 | const Node *Root; |
| 159 | clang::PrintingPolicy PrintPolicy; |
| 160 | |
| 161 | void print(llvm::raw_ostream &OS, const Node &N, int Indent) const; |
| 162 | friend llvm::raw_ostream &operator<<(llvm::raw_ostream &OS, |
| 163 | const SelectionTree &T) { |
| 164 | T.print(OS, N: T.root(), Indent: 1); |
| 165 | return OS; |
| 166 | } |
| 167 | }; |
| 168 | |
| 169 | } // namespace clangd |
| 170 | } // namespace clang |
| 171 | #endif |
| 172 | |