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 | |
49 | namespace clang { |
50 | namespace clangd { |
51 | |
52 | // A selection can partially or completely cover several AST nodes. |
53 | // The SelectionTree contains nodes that are covered, and their parents. |
54 | // SelectionTree does not contain all AST nodes, rather only: |
55 | // Decl, Stmt, TypeLoc, NestedNamespaceSpecifierLoc, CXXCtorInitializer. |
56 | // (These are the nodes with source ranges that fit in DynTypedNode). |
57 | // |
58 | // Usually commonAncestor() is the place to start: |
59 | // - it's the simplest answer to "what node is under the cursor" |
60 | // - the selected Expr (for example) can be found by walking up the parent |
61 | // chain and checking Node->ASTNode. |
62 | // - if you want to traverse the selected nodes, they are all under |
63 | // commonAncestor() in the tree. |
64 | // |
65 | // SelectionTree tries to behave sensibly in the presence of macros, but does |
66 | // not model any preprocessor concepts: the output is a subset of the AST. |
67 | // When a macro argument is specifically selected, only its first expansion is |
68 | // selected in the AST. (Returning a selection forest is unreasonably difficult |
69 | // for callers to handle correctly.) |
70 | // |
71 | // Comments, directives and whitespace are completely ignored. |
72 | // Semicolons are also ignored, as the AST generally does not model them well. |
73 | // |
74 | // The SelectionTree owns the Node structures, but the ASTNode attributes |
75 | // point back into the AST it was constructed with. |
76 | class SelectionTree { |
77 | public: |
78 | // Create selection trees for the given range, and pass them to Func. |
79 | // |
80 | // There may be multiple possible selection trees: |
81 | // - if the range is empty and borders two tokens, a tree for the right token |
82 | // and a tree for the left token will be yielded. |
83 | // - Func should return true on success (stop) and false on failure (continue) |
84 | // |
85 | // Always yields at least one tree. If no tokens are touched, it is empty. |
86 | static bool createEach(ASTContext &AST, const syntax::TokenBuffer &Tokens, |
87 | unsigned Begin, unsigned End, |
88 | llvm::function_ref<bool(SelectionTree)> Func); |
89 | |
90 | // Create a selection tree for the given range. |
91 | // |
92 | // Where ambiguous (range is empty and borders two tokens), prefer the token |
93 | // on the right. |
94 | static SelectionTree createRight(ASTContext &AST, |
95 | const syntax::TokenBuffer &Tokens, |
96 | unsigned Begin, unsigned End); |
97 | |
98 | // Copies are no good - contain pointers to other nodes. |
99 | SelectionTree(const SelectionTree &) = delete; |
100 | SelectionTree &operator=(const SelectionTree &) = delete; |
101 | // Moves are OK though - internal storage is pointer-stable when moved. |
102 | SelectionTree(SelectionTree &&) = default; |
103 | SelectionTree &operator=(SelectionTree &&) = default; |
104 | |
105 | // Describes to what extent an AST node is covered by the selection. |
106 | enum Selection : unsigned char { |
107 | // The AST node owns no characters covered by the selection. |
108 | // Note that characters owned by children don't count: |
109 | // if (x == 0) scream(); |
110 | // ^^^^^^ |
111 | // The IfStmt would be Unselected because all the selected characters are |
112 | // associated with its children. |
113 | // (Invisible nodes like ImplicitCastExpr are always unselected). |
114 | Unselected, |
115 | // The AST node owns selected characters, but is not completely covered. |
116 | Partial, |
117 | // The AST node owns characters, and is covered by the selection. |
118 | Complete, |
119 | }; |
120 | // An AST node that is implicated in the selection. |
121 | // (Either selected directly, or some descendant is selected). |
122 | struct Node { |
123 | // The parent within the selection tree. nullptr for TranslationUnitDecl. |
124 | Node *Parent; |
125 | // Direct children within the selection tree. |
126 | llvm::SmallVector<const Node *> Children; |
127 | // The corresponding node from the full AST. |
128 | DynTypedNode ASTNode; |
129 | // The extent to which this node is covered by the selection. |
130 | Selection Selected; |
131 | // Walk up the AST to get the lexical DeclContext of this Node, which is not |
132 | // the node itself. |
133 | const DeclContext &getDeclContext() const; |
134 | // Printable node kind, like "CXXRecordDecl" or "AutoTypeLoc". |
135 | std::string kind() const; |
136 | // If this node is a wrapper with no syntax (e.g. implicit cast), return |
137 | // its contents. (If multiple wrappers are present, unwraps all of them). |
138 | const Node &ignoreImplicit() const; |
139 | // If this node is inside a wrapper with no syntax (e.g. implicit cast), |
140 | // return that wrapper. (If multiple are present, unwraps all of them). |
141 | const Node &outerImplicit() const; |
142 | }; |
143 | // The most specific common ancestor of all the selected nodes. |
144 | // Returns nullptr if the common ancestor is the root. |
145 | // (This is to avoid accidentally traversing the TUDecl and thus preamble). |
146 | const Node *commonAncestor() const; |
147 | // The selection node corresponding to TranslationUnitDecl. |
148 | const Node &root() const { return *Root; } |
149 | |
150 | private: |
151 | // Creates a selection tree for the given range in the main file. |
152 | // The range includes bytes [Start, End). |
153 | SelectionTree(ASTContext &AST, const syntax::TokenBuffer &Tokens, |
154 | unsigned Start, unsigned End); |
155 | |
156 | std::deque<Node> Nodes; // Stable-pointer storage. |
157 | const Node *Root; |
158 | clang::PrintingPolicy PrintPolicy; |
159 | |
160 | void print(llvm::raw_ostream &OS, const Node &N, int Indent) const; |
161 | friend llvm::raw_ostream &operator<<(llvm::raw_ostream &OS, |
162 | const SelectionTree &T) { |
163 | T.print(OS, N: T.root(), Indent: 1); |
164 | return OS; |
165 | } |
166 | }; |
167 | |
168 | } // namespace clangd |
169 | } // namespace clang |
170 | #endif |
171 | |