1 | //===--- CodeComplete.h ------------------------------------------*- 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 | // Code completion provides suggestions for what the user might type next. |
10 | // After "std::string S; S." we might suggest members of std::string. |
11 | // Signature help describes the parameters of a function as you type them. |
12 | // |
13 | //===----------------------------------------------------------------------===// |
14 | |
15 | #ifndef LLVM_CLANG_TOOLS_EXTRA_CLANGD_CODECOMPLETE_H |
16 | #define |
17 | |
18 | #include "ASTSignals.h" |
19 | #include "Compiler.h" |
20 | #include "Protocol.h" |
21 | #include "Quality.h" |
22 | #include "index/Index.h" |
23 | #include "index/Symbol.h" |
24 | #include "index/SymbolOrigin.h" |
25 | #include "support/Markup.h" |
26 | #include "support/Path.h" |
27 | #include "clang/Sema/CodeCompleteConsumer.h" |
28 | #include "clang/Sema/CodeCompleteOptions.h" |
29 | #include "llvm/ADT/SmallVector.h" |
30 | #include "llvm/ADT/StringRef.h" |
31 | #include <functional> |
32 | #include <future> |
33 | #include <optional> |
34 | #include <utility> |
35 | |
36 | namespace clang { |
37 | class NamedDecl; |
38 | namespace clangd { |
39 | struct PreambleData; |
40 | struct CodeCompletion; |
41 | |
42 | struct CodeCompleteOptions { |
43 | /// Returns options that can be passed to clang's completion engine. |
44 | clang::CodeCompleteOptions getClangCompleteOpts() const; |
45 | |
46 | /// When true, completion items will contain expandable code snippets in |
47 | /// completion (e.g. `return ${1:expression}` or `foo(${1:int a}, ${2:int |
48 | /// b})). |
49 | bool EnableSnippets = false; |
50 | |
51 | /// Include results that are not legal completions in the current context. |
52 | /// For example, private members are usually inaccessible. |
53 | bool IncludeIneligibleResults = false; |
54 | |
55 | /// Combine overloads into a single completion item where possible. |
56 | /// If none, the implementation may choose an appropriate behavior. |
57 | /// (In practice, ClangdLSPServer enables bundling if the client claims |
58 | /// to supports signature help). |
59 | std::optional<bool> BundleOverloads; |
60 | |
61 | /// Limit the number of results returned (0 means no limit). |
62 | /// If more results are available, we set CompletionList.isIncomplete. |
63 | size_t Limit = 0; |
64 | |
65 | /// Whether to present doc comments as plain-text or markdown. |
66 | MarkupKind DocumentationFormat = MarkupKind::PlainText; |
67 | |
68 | enum IncludeInsertion { |
69 | IWYU, |
70 | NeverInsert, |
71 | } InsertIncludes = IncludeInsertion::IWYU; |
72 | |
73 | /// Whether include insertions for Objective-C code should use #import instead |
74 | /// of #include. |
75 | bool ImportInsertions = false; |
76 | |
77 | /// A visual indicator to prepend to the completion label to indicate whether |
78 | /// completion result would trigger an #include insertion or not. |
79 | struct IncludeInsertionIndicator { |
80 | std::string Insert = "•" ; |
81 | std::string NoInsert = " " ; |
82 | } IncludeIndicator; |
83 | |
84 | /// Expose origins of completion items in the label (for debugging). |
85 | bool ShowOrigins = false; |
86 | |
87 | // Populated internally by clangd, do not set. |
88 | /// If `Index` is set, it is used to augment the code completion |
89 | /// results. |
90 | /// FIXME(ioeric): we might want a better way to pass the index around inside |
91 | /// clangd. |
92 | const SymbolIndex *Index = nullptr; |
93 | |
94 | const ASTSignals *MainFileSignals = nullptr; |
95 | /// Include completions that require small corrections, e.g. change '.' to |
96 | /// '->' on member access etc. |
97 | bool IncludeFixIts = false; |
98 | |
99 | /// Whether to generate snippets for function arguments on code-completion. |
100 | /// Needs snippets to be enabled as well. |
101 | bool EnableFunctionArgSnippets = true; |
102 | |
103 | /// Whether to include index symbols that are not defined in the scopes |
104 | /// visible from the code completion point. This applies in contexts without |
105 | /// explicit scope qualifiers. |
106 | /// |
107 | /// Such completions can insert scope qualifiers. |
108 | bool AllScopes = false; |
109 | |
110 | /// Whether to use the clang parser, or fallback to text-based completion |
111 | /// (using identifiers in the current file and symbol indexes). |
112 | enum CodeCompletionParse { |
113 | /// Block until we can run the parser (e.g. preamble is built). |
114 | /// Return an error if this fails. |
115 | AlwaysParse, |
116 | /// Run the parser if inputs (preamble) are ready. |
117 | /// Otherwise, use text-based completion. |
118 | ParseIfReady, |
119 | /// Always use text-based completion. |
120 | NeverParse, |
121 | } RunParser = ParseIfReady; |
122 | |
123 | /// Callback invoked on all CompletionCandidate after they are scored and |
124 | /// before they are ranked (by -Score). Thus the results are yielded in |
125 | /// arbitrary order. |
126 | /// |
127 | /// This callbacks allows capturing various internal structures used by clangd |
128 | /// during code completion. Eg: Symbol quality and relevance signals. |
129 | std::function<void(const CodeCompletion &, const SymbolQualitySignals &, |
130 | const SymbolRelevanceSignals &, float Score)> |
131 | RecordCCResult; |
132 | |
133 | /// Model to use for ranking code completion candidates. |
134 | enum CodeCompletionRankingModel { |
135 | Heuristics, |
136 | DecisionForest, |
137 | }; |
138 | static const CodeCompletionRankingModel DefaultRankingModel; |
139 | CodeCompletionRankingModel RankingModel = DefaultRankingModel; |
140 | |
141 | /// Callback used to score a CompletionCandidate if DecisionForest ranking |
142 | /// model is enabled. |
143 | /// This allows us to inject experimental models and compare them with |
144 | /// baseline model using A/B testing. |
145 | std::function<DecisionForestScores( |
146 | const SymbolQualitySignals &, const SymbolRelevanceSignals &, float Base)> |
147 | DecisionForestScorer = &evaluateDecisionForest; |
148 | /// Weight for combining NameMatch and Prediction of DecisionForest. |
149 | /// CompletionScore is NameMatch * pow(Base, Prediction). |
150 | /// The optimal value of Base largely depends on the semantics of the model |
151 | /// and prediction score (e.g. algorithm used during training, number of |
152 | /// trees, etc.). Usually if the range of Prediction is [-20, 20] then a Base |
153 | /// in [1.2, 1.7] works fine. |
154 | /// Semantics: E.g. For Base = 1.3, if the Prediction score reduces by 2.6 |
155 | /// points then completion score reduces by 50% or 1.3^(-2.6). |
156 | float DecisionForestBase = 1.3f; |
157 | }; |
158 | |
159 | // Semi-structured representation of a code-complete suggestion for our C++ API. |
160 | // We don't use the LSP structures here (unlike most features) as we want |
161 | // to expose more data to allow for more precise testing and evaluation. |
162 | struct CodeCompletion { |
163 | // The unqualified name of the symbol or other completion item. |
164 | std::string Name; |
165 | // The name of the symbol for filtering and sorting purposes. Typically the |
166 | // same as `Name`, but may be different e.g. for ObjC methods, `Name` is the |
167 | // first selector fragment but the `FilterText` is the entire selector. |
168 | std::string FilterText; |
169 | // The scope qualifier for the symbol name. e.g. "ns1::ns2::" |
170 | // Empty for non-symbol completions. Not inserted, but may be displayed. |
171 | std::string Scope; |
172 | // Text that must be inserted before the name, and displayed (e.g. base::). |
173 | std::string RequiredQualifier; |
174 | // Details to be displayed following the name. Not inserted. |
175 | std::string Signature; |
176 | // Text to be inserted following the name, in snippet format. |
177 | std::string SnippetSuffix; |
178 | // Type to be displayed for this completion. |
179 | std::string ReturnType; |
180 | // The parsed documentation comment. |
181 | std::optional<markup::Document> Documentation; |
182 | CompletionItemKind Kind = CompletionItemKind::Missing; |
183 | // This completion item may represent several symbols that can be inserted in |
184 | // the same way, such as function overloads. In this case BundleSize > 1, and |
185 | // the following fields are summaries: |
186 | // - Signature is e.g. "(...)" for functions. |
187 | // - SnippetSuffix is similarly e.g. "(${0})". |
188 | // - ReturnType may be empty |
189 | // - Documentation may be from one symbol, or a combination of several |
190 | // Other fields should apply equally to all bundled completions. |
191 | unsigned BundleSize = 1; |
192 | SymbolOrigin Origin = SymbolOrigin::Unknown; |
193 | |
194 | struct IncludeCandidate { |
195 | // The header through which this symbol could be included. |
196 | // Quoted string as expected by an #include directive, e.g. "<memory>". |
197 | // Empty for non-symbol completions, or when not known. |
198 | std::string Header; |
199 | // Present if Header should be inserted to use this item. |
200 | std::optional<TextEdit> Insertion; |
201 | }; |
202 | // All possible include headers ranked by preference. By default, the first |
203 | // include is used. |
204 | // If we've bundled together overloads that have different sets of includes, |
205 | // thse includes may not be accurate for all of them. |
206 | llvm::SmallVector<IncludeCandidate, 1> Includes; |
207 | |
208 | /// Holds information about small corrections that needs to be done. Like |
209 | /// converting '->' to '.' on member access. |
210 | std::vector<TextEdit> FixIts; |
211 | |
212 | /// Holds the range of the token we are going to replace with this completion. |
213 | Range CompletionTokenRange; |
214 | |
215 | // Scores are used to rank completion items. |
216 | struct Scores { |
217 | // The score that items are ranked by. |
218 | float Total = 0.f; |
219 | |
220 | // The finalScore with the fuzzy name match score excluded. |
221 | // When filtering client-side, editors should calculate the new fuzzy score, |
222 | // whose scale is 0-1 (with 1 = prefix match, special case 2 = exact match), |
223 | // and recompute finalScore = fuzzyScore * symbolScore. |
224 | float ExcludingName = 0.f; |
225 | |
226 | // Component scores that contributed to the final score: |
227 | |
228 | // Quality describes how important we think this candidate is, |
229 | // independent of the query. |
230 | // e.g. symbols with lots of incoming references have higher quality. |
231 | float Quality = 0.f; |
232 | // Relevance describes how well this candidate matched the query. |
233 | // e.g. symbols from nearby files have higher relevance. |
234 | float Relevance = 0.f; |
235 | }; |
236 | Scores Score; |
237 | |
238 | /// Indicates if this item is deprecated. |
239 | bool Deprecated = false; |
240 | |
241 | // Serialize this to an LSP completion item. This is a lossy operation. |
242 | CompletionItem render(const CodeCompleteOptions &) const; |
243 | }; |
244 | raw_ostream &operator<<(raw_ostream &, const CodeCompletion &); |
245 | struct CodeCompleteResult { |
246 | std::vector<CodeCompletion> Completions; |
247 | bool HasMore = false; |
248 | CodeCompletionContext::Kind Context = CodeCompletionContext::CCC_Other; |
249 | // The text that is being directly completed. |
250 | // Example: foo.pb^ -> foo.push_back() |
251 | // ~~ |
252 | // Typically matches the textEdit.range of Completions, but not guaranteed to. |
253 | std::optional<Range> CompletionRange; |
254 | // Usually the source will be parsed with a real C++ parser. |
255 | // But heuristics may be used instead if e.g. the preamble is not ready. |
256 | bool RanParser = true; |
257 | }; |
258 | raw_ostream &operator<<(raw_ostream &, const CodeCompleteResult &); |
259 | |
260 | /// A speculative and asynchronous fuzzy find index request (based on cached |
261 | /// request) that can be sent before parsing sema. This would reduce completion |
262 | /// latency if the speculation succeeds. |
263 | struct SpeculativeFuzzyFind { |
264 | /// A cached request from past code completions. |
265 | /// Set by caller of `codeComplete()`. |
266 | std::optional<FuzzyFindRequest> CachedReq; |
267 | /// The actual request used by `codeComplete()`. |
268 | /// Set by `codeComplete()`. This can be used by callers to update cache. |
269 | std::optional<FuzzyFindRequest> NewReq; |
270 | /// The result is consumed by `codeComplete()` if speculation succeeded. |
271 | /// NOTE: the destructor will wait for the async call to finish. |
272 | std::future<std::pair<bool /*Incomplete*/, SymbolSlab>> Result; |
273 | }; |
274 | |
275 | /// Gets code completions at a specified \p Pos in \p FileName. |
276 | /// |
277 | /// If \p Preamble is nullptr, this runs code completion without compiling the |
278 | /// code. |
279 | /// |
280 | /// If \p SpecFuzzyFind is set, a speculative and asynchronous fuzzy find index |
281 | /// request (based on cached request) will be run before parsing sema. In case |
282 | /// the speculative result is used by code completion (e.g. speculation failed), |
283 | /// the speculative result is not consumed, and `SpecFuzzyFind` is only |
284 | /// destroyed when the async request finishes. |
285 | CodeCompleteResult codeComplete(PathRef FileName, Position Pos, |
286 | const PreambleData *Preamble, |
287 | const ParseInputs &ParseInput, |
288 | CodeCompleteOptions Opts, |
289 | SpeculativeFuzzyFind *SpecFuzzyFind = nullptr); |
290 | |
291 | /// Get signature help at a specified \p Pos in \p FileName. |
292 | SignatureHelp signatureHelp(PathRef FileName, Position Pos, |
293 | const PreambleData &Preamble, |
294 | const ParseInputs &ParseInput, |
295 | MarkupKind DocumentationFormat); |
296 | |
297 | // For index-based completion, we only consider: |
298 | // * symbols in namespaces or translation unit scopes (e.g. no class |
299 | // members, no locals) |
300 | // * enum constants (both scoped and unscoped) |
301 | // * primary templates (no specializations) |
302 | // For the other cases, we let Clang do the completion because it does not |
303 | // need any non-local information and it will be much better at following |
304 | // lookup rules. Other symbols still appear in the index for other purposes, |
305 | // like workspace/symbols or textDocument/definition, but are not used for code |
306 | // completion. |
307 | bool isIndexedForCodeCompletion(const NamedDecl &ND, ASTContext &ASTCtx); |
308 | |
309 | // Text immediately before the completion point that should be completed. |
310 | // This is heuristically derived from the source code, and is used when: |
311 | // - semantic analysis fails |
312 | // - semantic analysis may be slow, and we speculatively query the index |
313 | struct CompletionPrefix { |
314 | // The unqualified partial name. |
315 | // If there is none, begin() == end() == completion position. |
316 | llvm::StringRef Name; |
317 | // The spelled scope qualifier, such as Foo::. |
318 | // If there is none, begin() == end() == Name.begin(). |
319 | llvm::StringRef Qualifier; |
320 | }; |
321 | // Heuristically parses before Offset to determine what should be completed. |
322 | CompletionPrefix guessCompletionPrefix(llvm::StringRef Content, |
323 | unsigned Offset); |
324 | |
325 | // Whether it makes sense to complete at the point based on typed characters. |
326 | // For instance, we implicitly trigger at `a->^` but not at `a>^`. |
327 | bool allowImplicitCompletion(llvm::StringRef Content, unsigned Offset); |
328 | |
329 | } // namespace clangd |
330 | } // namespace clang |
331 | |
332 | #endif // LLVM_CLANG_TOOLS_EXTRA_CLANGD_CODECOMPLETE_H |
333 | |