1 | //===--- FindSymbols.cpp ------------------------------------*- 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 | #include "FindSymbols.h" |
9 | |
10 | #include "AST.h" |
11 | #include "FuzzyMatch.h" |
12 | #include "ParsedAST.h" |
13 | #include "Quality.h" |
14 | #include "SourceCode.h" |
15 | #include "index/Index.h" |
16 | #include "support/Logger.h" |
17 | #include "clang/AST/DeclTemplate.h" |
18 | #include "clang/Index/IndexSymbol.h" |
19 | #include "llvm/ADT/ArrayRef.h" |
20 | #include "llvm/ADT/STLExtras.h" |
21 | #include "llvm/ADT/SmallVector.h" |
22 | #include "llvm/ADT/StringRef.h" |
23 | #include <limits> |
24 | #include <optional> |
25 | #include <tuple> |
26 | |
27 | #define DEBUG_TYPE "FindSymbols" |
28 | |
29 | namespace clang { |
30 | namespace clangd { |
31 | |
32 | namespace { |
33 | using ScoredSymbolInfo = std::pair<float, SymbolInformation>; |
34 | struct ScoredSymbolGreater { |
35 | bool operator()(const ScoredSymbolInfo &L, const ScoredSymbolInfo &R) { |
36 | if (L.first != R.first) |
37 | return L.first > R.first; |
38 | return L.second.name < R.second.name; // Earlier name is better. |
39 | } |
40 | }; |
41 | |
42 | // Returns true if \p Query can be found as a sub-sequence inside \p Scope. |
43 | bool approximateScopeMatch(llvm::StringRef Scope, llvm::StringRef Query) { |
44 | assert(Scope.empty() || Scope.ends_with("::" )); |
45 | assert(Query.empty() || Query.ends_with("::" )); |
46 | while (!Scope.empty() && !Query.empty()) { |
47 | auto Colons = Scope.find(Str: "::" ); |
48 | assert(Colons != llvm::StringRef::npos); |
49 | |
50 | llvm::StringRef LeadingSpecifier = Scope.slice(Start: 0, End: Colons + 2); |
51 | Scope = Scope.slice(Start: Colons + 2, End: llvm::StringRef::npos); |
52 | Query.consume_front(Prefix: LeadingSpecifier); |
53 | } |
54 | return Query.empty(); |
55 | } |
56 | |
57 | } // namespace |
58 | |
59 | llvm::Expected<Location> indexToLSPLocation(const SymbolLocation &Loc, |
60 | llvm::StringRef TUPath) { |
61 | auto Path = URI::resolve(FileURI: Loc.FileURI, HintPath: TUPath); |
62 | if (!Path) |
63 | return error(Fmt: "Could not resolve path for file '{0}': {1}" , Vals: Loc.FileURI, |
64 | Vals: Path.takeError()); |
65 | Location L; |
66 | L.uri = URIForFile::canonicalize(AbsPath: *Path, TUPath); |
67 | Position Start, End; |
68 | Start.line = Loc.Start.line(); |
69 | Start.character = Loc.Start.column(); |
70 | End.line = Loc.End.line(); |
71 | End.character = Loc.End.column(); |
72 | L.range = {.start: Start, .end: End}; |
73 | return L; |
74 | } |
75 | |
76 | llvm::Expected<Location> symbolToLocation(const Symbol &Sym, |
77 | llvm::StringRef TUPath) { |
78 | // Prefer the definition over e.g. a function declaration in a header |
79 | return indexToLSPLocation( |
80 | Loc: Sym.Definition ? Sym.Definition : Sym.CanonicalDeclaration, TUPath); |
81 | } |
82 | |
83 | llvm::Expected<std::vector<SymbolInformation>> |
84 | getWorkspaceSymbols(llvm::StringRef Query, int Limit, |
85 | const SymbolIndex *const Index, llvm::StringRef HintPath) { |
86 | std::vector<SymbolInformation> Result; |
87 | if (!Index) |
88 | return Result; |
89 | |
90 | // Lookup for qualified names are performed as: |
91 | // - Exact namespaces are boosted by the index. |
92 | // - Approximate matches are (sub-scope match) included via AnyScope logic. |
93 | // - Non-matching namespaces (no sub-scope match) are post-filtered. |
94 | auto Names = splitQualifiedName(QName: Query); |
95 | |
96 | FuzzyFindRequest Req; |
97 | Req.Query = std::string(Names.second); |
98 | |
99 | // FuzzyFind doesn't want leading :: qualifier. |
100 | auto HasLeadingColons = Names.first.consume_front(Prefix: "::" ); |
101 | // Limit the query to specific namespace if it is fully-qualified. |
102 | Req.AnyScope = !HasLeadingColons; |
103 | // Boost symbols from desired namespace. |
104 | if (HasLeadingColons || !Names.first.empty()) |
105 | Req.Scopes = {std::string(Names.first)}; |
106 | if (Limit) { |
107 | Req.Limit = Limit; |
108 | // If we are boosting a specific scope allow more results to be retrieved, |
109 | // since some symbols from preferred namespaces might not make the cut. |
110 | if (Req.AnyScope && !Req.Scopes.empty()) |
111 | *Req.Limit *= 5; |
112 | } |
113 | TopN<ScoredSymbolInfo, ScoredSymbolGreater> Top( |
114 | Req.Limit ? *Req.Limit : std::numeric_limits<size_t>::max()); |
115 | FuzzyMatcher Filter(Req.Query); |
116 | |
117 | Index->fuzzyFind(Req, Callback: [HintPath, &Top, &Filter, AnyScope = Req.AnyScope, |
118 | ReqScope = Names.first](const Symbol &Sym) { |
119 | llvm::StringRef Scope = Sym.Scope; |
120 | // Fuzzyfind might return symbols from irrelevant namespaces if query was |
121 | // not fully-qualified, drop those. |
122 | if (AnyScope && !approximateScopeMatch(Scope, Query: ReqScope)) |
123 | return; |
124 | |
125 | auto Loc = symbolToLocation(Sym, TUPath: HintPath); |
126 | if (!Loc) { |
127 | log(Fmt: "Workspace symbols: {0}" , Vals: Loc.takeError()); |
128 | return; |
129 | } |
130 | |
131 | SymbolQualitySignals Quality; |
132 | Quality.merge(IndexResult: Sym); |
133 | SymbolRelevanceSignals Relevance; |
134 | Relevance.Name = Sym.Name; |
135 | Relevance.Query = SymbolRelevanceSignals::Generic; |
136 | // If symbol and request scopes do not match exactly, apply a penalty. |
137 | Relevance.InBaseClass = AnyScope && Scope != ReqScope; |
138 | if (auto NameMatch = Filter.match(Word: Sym.Name)) |
139 | Relevance.NameMatch = *NameMatch; |
140 | else { |
141 | log(Fmt: "Workspace symbol: {0} didn't match query {1}" , Vals: Sym.Name, |
142 | Vals: Filter.pattern()); |
143 | return; |
144 | } |
145 | Relevance.merge(IndexResult: Sym); |
146 | auto QualScore = Quality.evaluateHeuristics(); |
147 | auto RelScore = Relevance.evaluateHeuristics(); |
148 | auto Score = evaluateSymbolAndRelevance(SymbolQuality: QualScore, SymbolRelevance: RelScore); |
149 | dlog("FindSymbols: {0}{1} = {2}\n{3}{4}\n" , Sym.Scope, Sym.Name, Score, |
150 | Quality, Relevance); |
151 | |
152 | SymbolInformation Info; |
153 | Info.name = (Sym.Name + Sym.TemplateSpecializationArgs).str(); |
154 | Info.kind = indexSymbolKindToSymbolKind(Kind: Sym.SymInfo.Kind); |
155 | Info.location = *Loc; |
156 | Scope.consume_back(Suffix: "::" ); |
157 | Info.containerName = Scope.str(); |
158 | |
159 | // Exposed score excludes fuzzy-match component, for client-side re-ranking. |
160 | Info.score = Relevance.NameMatch > std::numeric_limits<float>::epsilon() |
161 | ? Score / Relevance.NameMatch |
162 | : QualScore; |
163 | Top.push(V: {Score, std::move(Info)}); |
164 | }); |
165 | for (auto &R : std::move(Top).items()) |
166 | Result.push_back(x: std::move(R.second)); |
167 | return Result; |
168 | } |
169 | |
170 | namespace { |
171 | std::string getSymbolName(ASTContext &Ctx, const NamedDecl &ND) { |
172 | // Print `MyClass(Category)` instead of `Category` and `MyClass()` instead |
173 | // of `anonymous`. |
174 | if (const auto *Container = dyn_cast<ObjCContainerDecl>(Val: &ND)) |
175 | return printObjCContainer(C: *Container); |
176 | // Differentiate between class and instance methods: print `-foo` instead of |
177 | // `foo` and `+sharedInstance` instead of `sharedInstance`. |
178 | if (const auto *Method = dyn_cast<ObjCMethodDecl>(Val: &ND)) { |
179 | std::string Name; |
180 | llvm::raw_string_ostream OS(Name); |
181 | |
182 | OS << (Method->isInstanceMethod() ? '-' : '+'); |
183 | Method->getSelector().print(OS); |
184 | |
185 | OS.flush(); |
186 | return Name; |
187 | } |
188 | return printName(Ctx, ND); |
189 | } |
190 | |
191 | std::string getSymbolDetail(ASTContext &Ctx, const NamedDecl &ND) { |
192 | PrintingPolicy P(Ctx.getPrintingPolicy()); |
193 | P.SuppressScope = true; |
194 | P.SuppressUnwrittenScope = true; |
195 | P.AnonymousTagLocations = false; |
196 | P.PolishForDeclaration = true; |
197 | std::string Detail; |
198 | llvm::raw_string_ostream OS(Detail); |
199 | if (ND.getDescribedTemplateParams()) { |
200 | OS << "template " ; |
201 | } |
202 | if (const auto *VD = dyn_cast<ValueDecl>(Val: &ND)) { |
203 | // FIXME: better printing for dependent type |
204 | if (isa<CXXConstructorDecl>(Val: VD)) { |
205 | std::string ConstructorType = VD->getType().getAsString(Policy: P); |
206 | // Print constructor type as "(int)" instead of "void (int)". |
207 | llvm::StringRef WithoutVoid = ConstructorType; |
208 | WithoutVoid.consume_front(Prefix: "void " ); |
209 | OS << WithoutVoid; |
210 | } else if (!isa<CXXDestructorDecl>(Val: VD)) { |
211 | VD->getType().print(OS, Policy: P); |
212 | } |
213 | } else if (const auto *TD = dyn_cast<TagDecl>(Val: &ND)) { |
214 | OS << TD->getKindName(); |
215 | } else if (isa<TypedefNameDecl>(Val: &ND)) { |
216 | OS << "type alias" ; |
217 | } else if (isa<ConceptDecl>(Val: &ND)) { |
218 | OS << "concept" ; |
219 | } |
220 | return std::move(OS.str()); |
221 | } |
222 | |
223 | std::optional<DocumentSymbol> declToSym(ASTContext &Ctx, const NamedDecl &ND) { |
224 | auto &SM = Ctx.getSourceManager(); |
225 | |
226 | SourceLocation BeginLoc = ND.getBeginLoc(); |
227 | SourceLocation EndLoc = ND.getEndLoc(); |
228 | const auto SymbolRange = |
229 | toHalfOpenFileRange(Mgr: SM, LangOpts: Ctx.getLangOpts(), R: {BeginLoc, EndLoc}); |
230 | if (!SymbolRange) |
231 | return std::nullopt; |
232 | |
233 | index::SymbolInfo SymInfo = index::getSymbolInfo(&ND); |
234 | // FIXME: This is not classifying constructors, destructors and operators |
235 | // correctly. |
236 | SymbolKind SK = indexSymbolKindToSymbolKind(Kind: SymInfo.Kind); |
237 | |
238 | DocumentSymbol SI; |
239 | SI.name = getSymbolName(Ctx, ND); |
240 | SI.kind = SK; |
241 | SI.deprecated = ND.isDeprecated(); |
242 | SI.range = Range{sourceLocToPosition(SM, SymbolRange->getBegin()), |
243 | sourceLocToPosition(SM, SymbolRange->getEnd())}; |
244 | SI.detail = getSymbolDetail(Ctx, ND); |
245 | |
246 | SourceLocation NameLoc = ND.getLocation(); |
247 | SourceLocation FallbackNameLoc; |
248 | if (NameLoc.isMacroID()) { |
249 | if (isSpelledInSource(Loc: NameLoc, SM)) { |
250 | // Prefer the spelling loc, but save the expansion loc as a fallback. |
251 | FallbackNameLoc = SM.getExpansionLoc(Loc: NameLoc); |
252 | NameLoc = SM.getSpellingLoc(Loc: NameLoc); |
253 | } else { |
254 | NameLoc = SM.getExpansionLoc(Loc: NameLoc); |
255 | } |
256 | } |
257 | auto ComputeSelectionRange = [&](SourceLocation L) -> Range { |
258 | Position NameBegin = sourceLocToPosition(SM, Loc: L); |
259 | Position NameEnd = sourceLocToPosition( |
260 | SM, Loc: Lexer::getLocForEndOfToken(Loc: L, Offset: 0, SM, LangOpts: Ctx.getLangOpts())); |
261 | return Range{.start: NameBegin, .end: NameEnd}; |
262 | }; |
263 | |
264 | SI.selectionRange = ComputeSelectionRange(NameLoc); |
265 | if (!SI.range.contains(Rng: SI.selectionRange) && FallbackNameLoc.isValid()) { |
266 | // 'selectionRange' must be contained in 'range'. In cases where clang |
267 | // reports unrelated ranges, we first try falling back to the expansion |
268 | // loc for the selection range. |
269 | SI.selectionRange = ComputeSelectionRange(FallbackNameLoc); |
270 | } |
271 | if (!SI.range.contains(Rng: SI.selectionRange)) { |
272 | // If the containment relationship still doesn't hold, throw away |
273 | // 'range' and use 'selectionRange' for both. |
274 | SI.range = SI.selectionRange; |
275 | } |
276 | return SI; |
277 | } |
278 | |
279 | /// A helper class to build an outline for the parse AST. It traverses the AST |
280 | /// directly instead of using RecursiveASTVisitor (RAV) for three main reasons: |
281 | /// - there is no way to keep RAV from traversing subtrees we are not |
282 | /// interested in. E.g. not traversing function locals or implicit template |
283 | /// instantiations. |
284 | /// - it's easier to combine results of recursive passes, |
285 | /// - visiting decls is actually simple, so we don't hit the complicated |
286 | /// cases that RAV mostly helps with (types, expressions, etc.) |
287 | class DocumentOutline { |
288 | // A DocumentSymbol we're constructing. |
289 | // We use this instead of DocumentSymbol directly so that we can keep track |
290 | // of the nodes we insert for macros. |
291 | class SymBuilder { |
292 | std::vector<SymBuilder> Children; |
293 | DocumentSymbol Symbol; // Symbol.children is empty, use Children instead. |
294 | // Macro expansions that this node or its parents are associated with. |
295 | // (Thus we will never create further children for these expansions). |
296 | llvm::SmallVector<SourceLocation> EnclosingMacroLoc; |
297 | |
298 | public: |
299 | DocumentSymbol build() && { |
300 | for (SymBuilder &C : Children) { |
301 | Symbol.children.push_back(x: std::move(C).build()); |
302 | // Expand range to ensure children nest properly, which editors expect. |
303 | // This can fix some edge-cases in the AST, but is vital for macros. |
304 | // A macro expansion "contains" AST node if it covers the node's primary |
305 | // location, but it may not span the node's whole range. |
306 | Symbol.range.start = |
307 | std::min(a: Symbol.range.start, b: Symbol.children.back().range.start); |
308 | Symbol.range.end = |
309 | std::max(a: Symbol.range.end, b: Symbol.children.back().range.end); |
310 | } |
311 | return std::move(Symbol); |
312 | } |
313 | |
314 | // Add a symbol as a child of the current one. |
315 | SymBuilder &addChild(DocumentSymbol S) { |
316 | Children.emplace_back(); |
317 | Children.back().EnclosingMacroLoc = EnclosingMacroLoc; |
318 | Children.back().Symbol = std::move(S); |
319 | return Children.back(); |
320 | } |
321 | |
322 | // Get an appropriate container for children of this symbol that were |
323 | // expanded from a macro (whose spelled name is Tok). |
324 | // |
325 | // This may return: |
326 | // - a macro symbol child of this (either new or previously created) |
327 | // - this scope itself, if it *is* the macro symbol or is nested within it |
328 | SymBuilder &inMacro(const syntax::Token &Tok, const SourceManager &SM, |
329 | std::optional<syntax::TokenBuffer::Expansion> Exp) { |
330 | if (llvm::is_contained(Range&: EnclosingMacroLoc, Element: Tok.location())) |
331 | return *this; |
332 | // If there's an existing child for this macro, we expect it to be last. |
333 | if (!Children.empty() && !Children.back().EnclosingMacroLoc.empty() && |
334 | Children.back().EnclosingMacroLoc.back() == Tok.location()) |
335 | return Children.back(); |
336 | |
337 | DocumentSymbol Sym; |
338 | Sym.name = Tok.text(SM).str(); |
339 | Sym.kind = SymbolKind::Null; // There's no suitable kind! |
340 | Sym.range = Sym.selectionRange = |
341 | halfOpenToRange(SM, R: Tok.range(SM).toCharRange(SM)); |
342 | |
343 | // FIXME: Exp is currently unavailable for nested expansions. |
344 | if (Exp) { |
345 | // Full range covers the macro args. |
346 | Sym.range = halfOpenToRange(SM, R: CharSourceRange::getCharRange( |
347 | B: Exp->Spelled.front().location(), |
348 | E: Exp->Spelled.back().endLocation())); |
349 | // Show macro args as detail. |
350 | llvm::raw_string_ostream OS(Sym.detail); |
351 | const syntax::Token *Prev = nullptr; |
352 | for (const auto &Tok : Exp->Spelled.drop_front()) { |
353 | // Don't dump arbitrarily long macro args. |
354 | if (OS.tell() > 80) { |
355 | OS << " ...)" ; |
356 | break; |
357 | } |
358 | if (Prev && Prev->endLocation() != Tok.location()) |
359 | OS << ' '; |
360 | OS << Tok.text(SM); |
361 | Prev = &Tok; |
362 | } |
363 | } |
364 | SymBuilder &Child = addChild(S: std::move(Sym)); |
365 | Child.EnclosingMacroLoc.push_back(Elt: Tok.location()); |
366 | return Child; |
367 | } |
368 | }; |
369 | |
370 | public: |
371 | DocumentOutline(ParsedAST &AST) : AST(AST) {} |
372 | |
373 | /// Builds the document outline for the generated AST. |
374 | std::vector<DocumentSymbol> build() { |
375 | SymBuilder Root; |
376 | for (auto &TopLevel : AST.getLocalTopLevelDecls()) |
377 | traverseDecl(D: TopLevel, Parent&: Root); |
378 | return std::move(std::move(Root).build().children); |
379 | } |
380 | |
381 | private: |
382 | enum class VisitKind { No, OnlyDecl, OnlyChildren, DeclAndChildren }; |
383 | |
384 | void traverseDecl(Decl *D, SymBuilder &Parent) { |
385 | // Skip symbols which do not originate from the main file. |
386 | if (!isInsideMainFile(Loc: D->getLocation(), SM: AST.getSourceManager())) |
387 | return; |
388 | |
389 | if (auto *Templ = llvm::dyn_cast<TemplateDecl>(Val: D)) { |
390 | // TemplatedDecl might be null, e.g. concepts. |
391 | if (auto *TD = Templ->getTemplatedDecl()) |
392 | D = TD; |
393 | } |
394 | |
395 | VisitKind Visit = shouldVisit(D); |
396 | if (Visit == VisitKind::No) |
397 | return; |
398 | |
399 | if (Visit == VisitKind::OnlyChildren) |
400 | return traverseChildren(D, Builder&: Parent); |
401 | |
402 | auto *ND = llvm::cast<NamedDecl>(Val: D); |
403 | auto Sym = declToSym(Ctx&: AST.getASTContext(), ND: *ND); |
404 | if (!Sym) |
405 | return; |
406 | SymBuilder &MacroParent = possibleMacroContainer(TargetLoc: D->getLocation(), Parent); |
407 | SymBuilder &Child = MacroParent.addChild(S: std::move(*Sym)); |
408 | |
409 | if (Visit == VisitKind::OnlyDecl) |
410 | return; |
411 | |
412 | assert(Visit == VisitKind::DeclAndChildren && "Unexpected VisitKind" ); |
413 | traverseChildren(ND, Child); |
414 | } |
415 | |
416 | // Determines where a decl should appear in the DocumentSymbol hierarchy. |
417 | // |
418 | // This is usually a direct child of the relevant AST parent. |
419 | // But we may also insert nodes for macros. Given: |
420 | // #define DECLARE_INT(V) int v; |
421 | // namespace a { DECLARE_INT(x) } |
422 | // We produce: |
423 | // Namespace a |
424 | // Macro DECLARE_INT(x) |
425 | // Variable x |
426 | // |
427 | // In the absence of macros, this method simply returns Parent. |
428 | // Otherwise it may return a macro expansion node instead. |
429 | // Each macro only has at most one node in the hierarchy, even if it expands |
430 | // to multiple decls. |
431 | SymBuilder &possibleMacroContainer(SourceLocation TargetLoc, |
432 | SymBuilder &Parent) { |
433 | const auto &SM = AST.getSourceManager(); |
434 | // Look at the path of macro-callers from the token to the main file. |
435 | // Note that along these paths we see the "outer" macro calls first. |
436 | SymBuilder *CurParent = &Parent; |
437 | for (SourceLocation Loc = TargetLoc; Loc.isMacroID(); |
438 | Loc = SM.getImmediateMacroCallerLoc(Loc)) { |
439 | // Find the virtual macro body that our token is being substituted into. |
440 | FileID MacroBody; |
441 | if (SM.isMacroArgExpansion(Loc)) { |
442 | // Loc is part of a macro arg being substituted into a macro body. |
443 | MacroBody = SM.getFileID(SpellingLoc: SM.getImmediateExpansionRange(Loc).getBegin()); |
444 | } else { |
445 | // Loc is already in the macro body. |
446 | MacroBody = SM.getFileID(SpellingLoc: Loc); |
447 | } |
448 | // The macro body is being substituted for a macro expansion, whose |
449 | // first token is the name of the macro. |
450 | SourceLocation MacroName = |
451 | SM.getSLocEntry(FID: MacroBody).getExpansion().getExpansionLocStart(); |
452 | // Only include the macro expansion in the outline if it was written |
453 | // directly in the main file, rather than expanded from another macro. |
454 | if (!MacroName.isValid() || !MacroName.isFileID()) |
455 | continue; |
456 | // All conditions satisfied, add the macro. |
457 | if (auto *Tok = AST.getTokens().spelledTokenAt(Loc: MacroName)) |
458 | CurParent = &CurParent->inMacro( |
459 | Tok: *Tok, SM, Exp: AST.getTokens().expansionStartingAt(Spelled: Tok)); |
460 | } |
461 | return *CurParent; |
462 | } |
463 | |
464 | void traverseChildren(Decl *D, SymBuilder &Builder) { |
465 | auto *Scope = llvm::dyn_cast<DeclContext>(Val: D); |
466 | if (!Scope) |
467 | return; |
468 | for (auto *C : Scope->decls()) |
469 | traverseDecl(D: C, Parent&: Builder); |
470 | } |
471 | |
472 | VisitKind shouldVisit(Decl *D) { |
473 | if (D->isImplicit()) |
474 | return VisitKind::No; |
475 | |
476 | if (llvm::isa<LinkageSpecDecl>(Val: D) || llvm::isa<ExportDecl>(Val: D)) |
477 | return VisitKind::OnlyChildren; |
478 | |
479 | if (!llvm::isa<NamedDecl>(Val: D)) |
480 | return VisitKind::No; |
481 | |
482 | if (auto *Func = llvm::dyn_cast<FunctionDecl>(Val: D)) { |
483 | // Some functions are implicit template instantiations, those should be |
484 | // ignored. |
485 | if (auto *Info = Func->getTemplateSpecializationInfo()) { |
486 | if (!Info->isExplicitInstantiationOrSpecialization()) |
487 | return VisitKind::No; |
488 | } |
489 | // Only visit the function itself, do not visit the children (i.e. |
490 | // function parameters, etc.) |
491 | return VisitKind::OnlyDecl; |
492 | } |
493 | // Handle template instantiations. We have three cases to consider: |
494 | // - explicit instantiations, e.g. 'template class std::vector<int>;' |
495 | // Visit the decl itself (it's present in the code), but not the |
496 | // children. |
497 | // - implicit instantiations, i.e. not written by the user. |
498 | // Do not visit at all, they are not present in the code. |
499 | // - explicit specialization, e.g. 'template <> class vector<bool> {};' |
500 | // Visit both the decl and its children, both are written in the code. |
501 | if (auto *TemplSpec = llvm::dyn_cast<ClassTemplateSpecializationDecl>(Val: D)) { |
502 | if (TemplSpec->isExplicitInstantiationOrSpecialization()) |
503 | return TemplSpec->isExplicitSpecialization() |
504 | ? VisitKind::DeclAndChildren |
505 | : VisitKind::OnlyDecl; |
506 | return VisitKind::No; |
507 | } |
508 | if (auto *TemplSpec = llvm::dyn_cast<VarTemplateSpecializationDecl>(Val: D)) { |
509 | if (TemplSpec->isExplicitInstantiationOrSpecialization()) |
510 | return TemplSpec->isExplicitSpecialization() |
511 | ? VisitKind::DeclAndChildren |
512 | : VisitKind::OnlyDecl; |
513 | return VisitKind::No; |
514 | } |
515 | // For all other cases, visit both the children and the decl. |
516 | return VisitKind::DeclAndChildren; |
517 | } |
518 | |
519 | ParsedAST &AST; |
520 | }; |
521 | |
522 | struct PragmaMarkSymbol { |
523 | DocumentSymbol DocSym; |
524 | bool IsGroup; |
525 | }; |
526 | |
527 | /// Merge in `PragmaMarkSymbols`, sorted ascending by range, into the given |
528 | /// `DocumentSymbol` tree. |
529 | void mergePragmas(DocumentSymbol &Root, ArrayRef<PragmaMarkSymbol> Pragmas) { |
530 | while (!Pragmas.empty()) { |
531 | // We'll figure out where the Pragmas.front() should go. |
532 | PragmaMarkSymbol P = std::move(Pragmas.front()); |
533 | Pragmas = Pragmas.drop_front(); |
534 | DocumentSymbol *Cur = &Root; |
535 | while (Cur->range.contains(Rng: P.DocSym.range)) { |
536 | bool Swapped = false; |
537 | for (auto &C : Cur->children) { |
538 | // We assume at most 1 child can contain the pragma (as pragmas are on |
539 | // a single line, and children have disjoint ranges). |
540 | if (C.range.contains(Rng: P.DocSym.range)) { |
541 | Cur = &C; |
542 | Swapped = true; |
543 | break; |
544 | } |
545 | } |
546 | // Cur is the parent of P since none of the children contain P. |
547 | if (!Swapped) |
548 | break; |
549 | } |
550 | // Pragma isn't a group so we can just insert it and we are done. |
551 | if (!P.IsGroup) { |
552 | Cur->children.emplace_back(args: std::move(P.DocSym)); |
553 | continue; |
554 | } |
555 | // Pragma is a group, so we need to figure out where it terminates: |
556 | // - If the next Pragma is not contained in Cur, P owns all of its |
557 | // parent's children which occur after P. |
558 | // - If the next pragma is contained in Cur but actually belongs to one |
559 | // of the parent's children, we temporarily skip over it and look at |
560 | // the next pragma to decide where we end. |
561 | // - Otherwise nest all of its parent's children which occur after P but |
562 | // before the next pragma. |
563 | bool TerminatedByNextPragma = false; |
564 | for (auto &NextPragma : Pragmas) { |
565 | // If we hit a pragma outside of Cur, the rest will be outside as well. |
566 | if (!Cur->range.contains(Rng: NextPragma.DocSym.range)) |
567 | break; |
568 | |
569 | // NextPragma cannot terminate P if it is nested inside a child, look for |
570 | // the next one. |
571 | if (llvm::any_of(Range&: Cur->children, P: [&NextPragma](const auto &Child) { |
572 | return Child.range.contains(NextPragma.DocSym.range); |
573 | })) |
574 | continue; |
575 | |
576 | // Pragma owns all the children between P and NextPragma |
577 | auto It = llvm::partition(Range&: Cur->children, |
578 | P: [&P, &NextPragma](const auto &S) -> bool { |
579 | return !(P.DocSym.range < S.range && |
580 | S.range < NextPragma.DocSym.range); |
581 | }); |
582 | P.DocSym.children.assign(first: make_move_iterator(i: It), |
583 | last: make_move_iterator(i: Cur->children.end())); |
584 | Cur->children.erase(first: It, last: Cur->children.end()); |
585 | TerminatedByNextPragma = true; |
586 | break; |
587 | } |
588 | if (!TerminatedByNextPragma) { |
589 | // P is terminated by the end of current symbol, hence it owns all the |
590 | // children after P. |
591 | auto It = llvm::partition(Range&: Cur->children, P: [&P](const auto &S) -> bool { |
592 | return !(P.DocSym.range < S.range); |
593 | }); |
594 | P.DocSym.children.assign(first: make_move_iterator(i: It), |
595 | last: make_move_iterator(i: Cur->children.end())); |
596 | Cur->children.erase(first: It, last: Cur->children.end()); |
597 | } |
598 | // Update the range for P to cover children and append to Cur. |
599 | for (DocumentSymbol &Sym : P.DocSym.children) |
600 | unionRanges(A&: P.DocSym.range, B: Sym.range); |
601 | Cur->children.emplace_back(args: std::move(P.DocSym)); |
602 | } |
603 | } |
604 | |
605 | PragmaMarkSymbol markToSymbol(const PragmaMark &P) { |
606 | StringRef Name = StringRef(P.Trivia).trim(); |
607 | bool IsGroup = false; |
608 | // "-\s+<group name>" or "<name>" after an initial trim. The former is |
609 | // considered a group, the latter just a mark. Like Xcode, we don't consider |
610 | // `-Foo` to be a group (space(s) after the `-` is required). |
611 | // |
612 | // We need to include a name here, otherwise editors won't properly render the |
613 | // symbol. |
614 | StringRef MaybeGroupName = Name; |
615 | if (MaybeGroupName.consume_front(Prefix: "-" ) && |
616 | (MaybeGroupName.ltrim() != MaybeGroupName || MaybeGroupName.empty())) { |
617 | Name = MaybeGroupName.empty() ? "(unnamed group)" : MaybeGroupName.ltrim(); |
618 | IsGroup = true; |
619 | } else if (Name.empty()) { |
620 | Name = "(unnamed mark)" ; |
621 | } |
622 | DocumentSymbol Sym; |
623 | Sym.name = Name.str(); |
624 | Sym.kind = SymbolKind::File; |
625 | Sym.range = P.Rng; |
626 | Sym.selectionRange = P.Rng; |
627 | return {.DocSym: Sym, .IsGroup: IsGroup}; |
628 | } |
629 | |
630 | std::vector<DocumentSymbol> collectDocSymbols(ParsedAST &AST) { |
631 | std::vector<DocumentSymbol> Syms = DocumentOutline(AST).build(); |
632 | |
633 | const auto &PragmaMarks = AST.getMarks(); |
634 | if (PragmaMarks.empty()) |
635 | return Syms; |
636 | |
637 | std::vector<PragmaMarkSymbol> Pragmas; |
638 | Pragmas.reserve(n: PragmaMarks.size()); |
639 | for (const auto &P : PragmaMarks) |
640 | Pragmas.push_back(x: markToSymbol(P)); |
641 | Range EntireFile = { |
642 | .start: {.line: 0, .character: 0}, |
643 | .end: {.line: std::numeric_limits<int>::max(), .character: std::numeric_limits<int>::max()}}; |
644 | DocumentSymbol Root; |
645 | Root.children = std::move(Syms); |
646 | Root.range = EntireFile; |
647 | mergePragmas(Root, Pragmas: llvm::ArrayRef(Pragmas)); |
648 | return Root.children; |
649 | } |
650 | |
651 | } // namespace |
652 | |
653 | llvm::Expected<std::vector<DocumentSymbol>> getDocumentSymbols(ParsedAST &AST) { |
654 | return collectDocSymbols(AST); |
655 | } |
656 | |
657 | } // namespace clangd |
658 | } // namespace clang |
659 | |