| 1 | //===--- CodeComplete.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 | // |
| 9 | // Code completion has several moving parts: |
| 10 | // - AST-based completions are provided using the completion hooks in Sema. |
| 11 | // - external completions are retrieved from the index (using hints from Sema) |
| 12 | // - the two sources overlap, and must be merged and overloads bundled |
| 13 | // - results must be scored and ranked (see Quality.h) before rendering |
| 14 | // |
| 15 | // Signature help works in a similar way as code completion, but it is simpler: |
| 16 | // it's purely AST-based, and there are few candidates. |
| 17 | // |
| 18 | //===----------------------------------------------------------------------===// |
| 19 | |
| 20 | #include "CodeComplete.h" |
| 21 | #include "AST.h" |
| 22 | #include "CodeCompletionStrings.h" |
| 23 | #include "Compiler.h" |
| 24 | #include "Config.h" |
| 25 | #include "ExpectedTypes.h" |
| 26 | #include "Feature.h" |
| 27 | #include "FileDistance.h" |
| 28 | #include "FuzzyMatch.h" |
| 29 | #include "Headers.h" |
| 30 | #include "Hover.h" |
| 31 | #include "Preamble.h" |
| 32 | #include "Protocol.h" |
| 33 | #include "Quality.h" |
| 34 | #include "SourceCode.h" |
| 35 | #include "URI.h" |
| 36 | #include "index/Index.h" |
| 37 | #include "index/Symbol.h" |
| 38 | #include "index/SymbolOrigin.h" |
| 39 | #include "support/Logger.h" |
| 40 | #include "support/Markup.h" |
| 41 | #include "support/Threading.h" |
| 42 | #include "support/ThreadsafeFS.h" |
| 43 | #include "support/Trace.h" |
| 44 | #include "clang/AST/Decl.h" |
| 45 | #include "clang/AST/DeclBase.h" |
| 46 | #include "clang/Basic/CharInfo.h" |
| 47 | #include "clang/Basic/LangOptions.h" |
| 48 | #include "clang/Basic/SourceLocation.h" |
| 49 | #include "clang/Basic/TokenKinds.h" |
| 50 | #include "clang/Format/Format.h" |
| 51 | #include "clang/Frontend/CompilerInstance.h" |
| 52 | #include "clang/Frontend/FrontendActions.h" |
| 53 | #include "clang/Lex/ExternalPreprocessorSource.h" |
| 54 | #include "clang/Lex/Lexer.h" |
| 55 | #include "clang/Lex/Preprocessor.h" |
| 56 | #include "clang/Lex/PreprocessorOptions.h" |
| 57 | #include "clang/Sema/CodeCompleteConsumer.h" |
| 58 | #include "clang/Sema/DeclSpec.h" |
| 59 | #include "clang/Sema/Sema.h" |
| 60 | #include "llvm/ADT/ArrayRef.h" |
| 61 | #include "llvm/ADT/SmallVector.h" |
| 62 | #include "llvm/ADT/StringExtras.h" |
| 63 | #include "llvm/ADT/StringRef.h" |
| 64 | #include "llvm/Support/Casting.h" |
| 65 | #include "llvm/Support/Compiler.h" |
| 66 | #include "llvm/Support/Debug.h" |
| 67 | #include "llvm/Support/Error.h" |
| 68 | #include "llvm/Support/FormatVariadic.h" |
| 69 | #include "llvm/Support/ScopedPrinter.h" |
| 70 | #include <algorithm> |
| 71 | #include <iterator> |
| 72 | #include <limits> |
| 73 | #include <optional> |
| 74 | #include <utility> |
| 75 | |
| 76 | // We log detailed candidate here if you run with -debug-only=codecomplete. |
| 77 | #define DEBUG_TYPE "CodeComplete" |
| 78 | |
| 79 | namespace clang { |
| 80 | namespace clangd { |
| 81 | |
| 82 | #if CLANGD_DECISION_FOREST |
| 83 | const CodeCompleteOptions::CodeCompletionRankingModel |
| 84 | CodeCompleteOptions::DefaultRankingModel = |
| 85 | CodeCompleteOptions::DecisionForest; |
| 86 | #else |
| 87 | const CodeCompleteOptions::CodeCompletionRankingModel |
| 88 | CodeCompleteOptions::DefaultRankingModel = CodeCompleteOptions::Heuristics; |
| 89 | #endif |
| 90 | |
| 91 | namespace { |
| 92 | |
| 93 | // Note: changes to this function should also be reflected in the |
| 94 | // CodeCompletionResult overload where appropriate. |
| 95 | CompletionItemKind |
| 96 | toCompletionItemKind(index::SymbolKind Kind, |
| 97 | const llvm::StringRef *Signature = nullptr) { |
| 98 | using SK = index::SymbolKind; |
| 99 | switch (Kind) { |
| 100 | case SK::Unknown: |
| 101 | return CompletionItemKind::Missing; |
| 102 | case SK::Module: |
| 103 | case SK::Namespace: |
| 104 | case SK::NamespaceAlias: |
| 105 | return CompletionItemKind::Module; |
| 106 | case SK::Macro: |
| 107 | // Use macro signature (if provided) to tell apart function-like and |
| 108 | // object-like macros. |
| 109 | return Signature && Signature->contains(C: '(') ? CompletionItemKind::Function |
| 110 | : CompletionItemKind::Constant; |
| 111 | case SK::Enum: |
| 112 | return CompletionItemKind::Enum; |
| 113 | case SK::Struct: |
| 114 | return CompletionItemKind::Struct; |
| 115 | case SK::Class: |
| 116 | case SK::Extension: |
| 117 | case SK::Union: |
| 118 | return CompletionItemKind::Class; |
| 119 | case SK::Protocol: |
| 120 | // Use interface instead of class for differentiation of classes and |
| 121 | // protocols with the same name (e.g. @interface NSObject vs. @protocol |
| 122 | // NSObject). |
| 123 | return CompletionItemKind::Interface; |
| 124 | case SK::TypeAlias: |
| 125 | // We use the same kind as the VSCode C++ extension. |
| 126 | // FIXME: pick a better option when we have one. |
| 127 | return CompletionItemKind::Interface; |
| 128 | case SK::Using: |
| 129 | return CompletionItemKind::Reference; |
| 130 | case SK::Function: |
| 131 | case SK::ConversionFunction: |
| 132 | return CompletionItemKind::Function; |
| 133 | case SK::Variable: |
| 134 | case SK::Parameter: |
| 135 | case SK::NonTypeTemplateParm: |
| 136 | return CompletionItemKind::Variable; |
| 137 | case SK::Field: |
| 138 | return CompletionItemKind::Field; |
| 139 | case SK::EnumConstant: |
| 140 | return CompletionItemKind::EnumMember; |
| 141 | case SK::InstanceMethod: |
| 142 | case SK::ClassMethod: |
| 143 | case SK::StaticMethod: |
| 144 | case SK::Destructor: |
| 145 | return CompletionItemKind::Method; |
| 146 | case SK::InstanceProperty: |
| 147 | case SK::ClassProperty: |
| 148 | case SK::StaticProperty: |
| 149 | return CompletionItemKind::Property; |
| 150 | case SK::Constructor: |
| 151 | return CompletionItemKind::Constructor; |
| 152 | case SK::TemplateTypeParm: |
| 153 | case SK::TemplateTemplateParm: |
| 154 | return CompletionItemKind::TypeParameter; |
| 155 | case SK::Concept: |
| 156 | return CompletionItemKind::Interface; |
| 157 | } |
| 158 | llvm_unreachable("Unhandled clang::index::SymbolKind." ); |
| 159 | } |
| 160 | |
| 161 | // Note: changes to this function should also be reflected in the |
| 162 | // index::SymbolKind overload where appropriate. |
| 163 | CompletionItemKind toCompletionItemKind(const CodeCompletionResult &Res, |
| 164 | CodeCompletionContext::Kind CtxKind) { |
| 165 | if (Res.Declaration) |
| 166 | return toCompletionItemKind(index::getSymbolInfo(Res.Declaration).Kind); |
| 167 | if (CtxKind == CodeCompletionContext::CCC_IncludedFile) |
| 168 | return CompletionItemKind::File; |
| 169 | switch (Res.Kind) { |
| 170 | case CodeCompletionResult::RK_Declaration: |
| 171 | llvm_unreachable("RK_Declaration without Decl" ); |
| 172 | case CodeCompletionResult::RK_Keyword: |
| 173 | return CompletionItemKind::Keyword; |
| 174 | case CodeCompletionResult::RK_Macro: |
| 175 | // There is no 'Macro' kind in LSP. |
| 176 | // Avoid using 'Text' to avoid confusion with client-side word-based |
| 177 | // completion proposals. |
| 178 | return Res.MacroDefInfo && Res.MacroDefInfo->isFunctionLike() |
| 179 | ? CompletionItemKind::Function |
| 180 | : CompletionItemKind::Constant; |
| 181 | case CodeCompletionResult::RK_Pattern: |
| 182 | return CompletionItemKind::Snippet; |
| 183 | } |
| 184 | llvm_unreachable("Unhandled CodeCompletionResult::ResultKind." ); |
| 185 | } |
| 186 | |
| 187 | // FIXME: find a home for this (that can depend on both markup and Protocol). |
| 188 | MarkupContent renderDoc(const markup::Document &Doc, MarkupKind Kind) { |
| 189 | MarkupContent Result; |
| 190 | Result.kind = Kind; |
| 191 | switch (Kind) { |
| 192 | case MarkupKind::PlainText: |
| 193 | Result.value.append(str: Doc.asPlainText()); |
| 194 | break; |
| 195 | case MarkupKind::Markdown: |
| 196 | Result.value.append(str: Doc.asMarkdown()); |
| 197 | break; |
| 198 | } |
| 199 | return Result; |
| 200 | } |
| 201 | |
| 202 | Symbol::IncludeDirective insertionDirective(const CodeCompleteOptions &Opts) { |
| 203 | if (!Opts.ImportInsertions || !Opts.MainFileSignals) |
| 204 | return Symbol::IncludeDirective::Include; |
| 205 | return Opts.MainFileSignals->InsertionDirective; |
| 206 | } |
| 207 | |
| 208 | // Identifier code completion result. |
| 209 | struct RawIdentifier { |
| 210 | llvm::StringRef Name; |
| 211 | unsigned References; // # of usages in file. |
| 212 | }; |
| 213 | |
| 214 | /// A code completion result, in clang-native form. |
| 215 | /// It may be promoted to a CompletionItem if it's among the top-ranked results. |
| 216 | struct CompletionCandidate { |
| 217 | llvm::StringRef Name; // Used for filtering and sorting. |
| 218 | // We may have a result from Sema, from the index, or both. |
| 219 | const CodeCompletionResult *SemaResult = nullptr; |
| 220 | const Symbol *IndexResult = nullptr; |
| 221 | const RawIdentifier *IdentifierResult = nullptr; |
| 222 | llvm::SmallVector<SymbolInclude, 1> RankedIncludeHeaders; |
| 223 | |
| 224 | // Returns a token identifying the overload set this is part of. |
| 225 | // 0 indicates it's not part of any overload set. |
| 226 | size_t overloadSet(const CodeCompleteOptions &Opts, llvm::StringRef FileName, |
| 227 | IncludeInserter *Inserter, |
| 228 | CodeCompletionContext::Kind CCContextKind) const { |
| 229 | if (!Opts.BundleOverloads.value_or(u: false)) |
| 230 | return 0; |
| 231 | |
| 232 | // Depending on the index implementation, we can see different header |
| 233 | // strings (literal or URI) mapping to the same file. We still want to |
| 234 | // bundle those, so we must resolve the header to be included here. |
| 235 | std::string ; |
| 236 | if (Inserter) { |
| 237 | if (auto = headerToInsertIfAllowed(Opts, ContextKind: CCContextKind)) { |
| 238 | if (auto = toHeaderFile(Header: *Header, HintPath: FileName)) { |
| 239 | if (auto Spelled = |
| 240 | Inserter->calculateIncludePath(InsertedHeader: *HeaderFile, IncludingFile: FileName)) |
| 241 | HeaderForHash = *Spelled; |
| 242 | } else { |
| 243 | vlog(Fmt: "Code completion header path manipulation failed {0}" , |
| 244 | Vals: HeaderFile.takeError()); |
| 245 | } |
| 246 | } |
| 247 | } |
| 248 | |
| 249 | llvm::SmallString<256> Scratch; |
| 250 | if (IndexResult) { |
| 251 | switch (IndexResult->SymInfo.Kind) { |
| 252 | case index::SymbolKind::ClassMethod: |
| 253 | case index::SymbolKind::InstanceMethod: |
| 254 | case index::SymbolKind::StaticMethod: |
| 255 | #ifndef NDEBUG |
| 256 | llvm_unreachable("Don't expect members from index in code completion" ); |
| 257 | #else |
| 258 | [[fallthrough]]; |
| 259 | #endif |
| 260 | case index::SymbolKind::Function: |
| 261 | // We can't group overloads together that need different #includes. |
| 262 | // This could break #include insertion. |
| 263 | return llvm::hash_combine( |
| 264 | args: (IndexResult->Scope + IndexResult->Name).toStringRef(Out&: Scratch), |
| 265 | args: HeaderForHash); |
| 266 | default: |
| 267 | return 0; |
| 268 | } |
| 269 | } |
| 270 | if (SemaResult) { |
| 271 | // We need to make sure we're consistent with the IndexResult case! |
| 272 | const NamedDecl *D = SemaResult->Declaration; |
| 273 | if (!D || !D->isFunctionOrFunctionTemplate()) |
| 274 | return 0; |
| 275 | { |
| 276 | llvm::raw_svector_ostream OS(Scratch); |
| 277 | D->printQualifiedName(OS); |
| 278 | } |
| 279 | return llvm::hash_combine(args: Scratch, args: HeaderForHash); |
| 280 | } |
| 281 | assert(IdentifierResult); |
| 282 | return 0; |
| 283 | } |
| 284 | |
| 285 | bool contextAllowsHeaderInsertion(CodeCompletionContext::Kind Kind) const { |
| 286 | // Explicitly disable insertions for forward declarations since they don't |
| 287 | // reference the declaration. |
| 288 | if (Kind == CodeCompletionContext::CCC_ObjCClassForwardDecl) |
| 289 | return false; |
| 290 | return true; |
| 291 | } |
| 292 | |
| 293 | // The best header to include if include insertion is allowed. |
| 294 | std::optional<llvm::StringRef> |
| 295 | headerToInsertIfAllowed(const CodeCompleteOptions &Opts, |
| 296 | CodeCompletionContext::Kind ContextKind) const { |
| 297 | if (Opts.InsertIncludes == Config::HeaderInsertionPolicy::NeverInsert || |
| 298 | RankedIncludeHeaders.empty() || |
| 299 | !contextAllowsHeaderInsertion(Kind: ContextKind)) |
| 300 | return std::nullopt; |
| 301 | if (SemaResult && SemaResult->Declaration) { |
| 302 | // Avoid inserting new #include if the declaration is found in the current |
| 303 | // file e.g. the symbol is forward declared. |
| 304 | auto &SM = SemaResult->Declaration->getASTContext().getSourceManager(); |
| 305 | for (const Decl *RD : SemaResult->Declaration->redecls()) |
| 306 | if (SM.isInMainFile(SM.getExpansionLoc(RD->getBeginLoc()))) |
| 307 | return std::nullopt; |
| 308 | } |
| 309 | Symbol::IncludeDirective Directive = insertionDirective(Opts); |
| 310 | for (const auto &Inc : RankedIncludeHeaders) |
| 311 | if ((Inc.Directive & Directive) != 0) |
| 312 | return Inc.Header; |
| 313 | return std::nullopt; |
| 314 | } |
| 315 | |
| 316 | using Bundle = llvm::SmallVector<CompletionCandidate, 4>; |
| 317 | }; |
| 318 | using ScoredBundle = |
| 319 | std::pair<CompletionCandidate::Bundle, CodeCompletion::Scores>; |
| 320 | struct ScoredBundleGreater { |
| 321 | bool operator()(const ScoredBundle &L, const ScoredBundle &R) { |
| 322 | if (L.second.Total != R.second.Total) |
| 323 | return L.second.Total > R.second.Total; |
| 324 | return L.first.front().Name < |
| 325 | R.first.front().Name; // Earlier name is better. |
| 326 | } |
| 327 | }; |
| 328 | |
| 329 | // Remove the first template argument from Signature. |
| 330 | // If Signature only contains a single argument an empty string is returned. |
| 331 | std::string removeFirstTemplateArg(llvm::StringRef Signature) { |
| 332 | auto Rest = Signature.split(Separator: "," ).second; |
| 333 | if (Rest.empty()) |
| 334 | return "" ; |
| 335 | return ("<" + Rest.ltrim()).str(); |
| 336 | } |
| 337 | |
| 338 | // Assembles a code completion out of a bundle of >=1 completion candidates. |
| 339 | // Many of the expensive strings are only computed at this point, once we know |
| 340 | // the candidate bundle is going to be returned. |
| 341 | // |
| 342 | // Many fields are the same for all candidates in a bundle (e.g. name), and are |
| 343 | // computed from the first candidate, in the constructor. |
| 344 | // Others vary per candidate, so add() must be called for remaining candidates. |
| 345 | struct CodeCompletionBuilder { |
| 346 | CodeCompletionBuilder(ASTContext *ASTCtx, const CompletionCandidate &C, |
| 347 | CodeCompletionString *SemaCCS, |
| 348 | llvm::ArrayRef<std::string> AccessibleScopes, |
| 349 | const IncludeInserter &Includes, |
| 350 | llvm::StringRef FileName, |
| 351 | CodeCompletionContext::Kind ContextKind, |
| 352 | const CodeCompleteOptions &Opts, |
| 353 | bool IsUsingDeclaration, tok::TokenKind NextTokenKind) |
| 354 | : ASTCtx(ASTCtx), ArgumentLists(Opts.ArgumentLists), |
| 355 | IsUsingDeclaration(IsUsingDeclaration), NextTokenKind(NextTokenKind) { |
| 356 | Completion.Deprecated = true; // cleared by any non-deprecated overload. |
| 357 | add(C, SemaCCS, ContextKind); |
| 358 | if (C.SemaResult) { |
| 359 | assert(ASTCtx); |
| 360 | Completion.Origin |= SymbolOrigin::AST; |
| 361 | Completion.Name = std::string(llvm::StringRef(SemaCCS->getTypedText())); |
| 362 | Completion.FilterText = SemaCCS->getAllTypedText(); |
| 363 | if (Completion.Scope.empty()) { |
| 364 | if ((C.SemaResult->Kind == CodeCompletionResult::RK_Declaration) || |
| 365 | (C.SemaResult->Kind == CodeCompletionResult::RK_Pattern)) |
| 366 | if (const auto *D = C.SemaResult->getDeclaration()) |
| 367 | if (const auto *ND = dyn_cast<NamedDecl>(Val: D)) |
| 368 | Completion.Scope = std::string( |
| 369 | splitQualifiedName(QName: printQualifiedName(ND: *ND)).first); |
| 370 | } |
| 371 | Completion.Kind = toCompletionItemKind(Res: *C.SemaResult, CtxKind: ContextKind); |
| 372 | // Sema could provide more info on whether the completion was a file or |
| 373 | // folder. |
| 374 | if (Completion.Kind == CompletionItemKind::File && |
| 375 | Completion.Name.back() == '/') |
| 376 | Completion.Kind = CompletionItemKind::Folder; |
| 377 | for (const auto &FixIt : C.SemaResult->FixIts) { |
| 378 | Completion.FixIts.push_back(x: toTextEdit( |
| 379 | FixIt, M: ASTCtx->getSourceManager(), L: ASTCtx->getLangOpts())); |
| 380 | } |
| 381 | llvm::sort(C&: Completion.FixIts, Comp: [](const TextEdit &X, const TextEdit &Y) { |
| 382 | return std::tie(args: X.range.start.line, args: X.range.start.character) < |
| 383 | std::tie(args: Y.range.start.line, args: Y.range.start.character); |
| 384 | }); |
| 385 | } |
| 386 | if (C.IndexResult) { |
| 387 | Completion.Origin |= C.IndexResult->Origin; |
| 388 | if (Completion.Scope.empty()) |
| 389 | Completion.Scope = std::string(C.IndexResult->Scope); |
| 390 | if (Completion.Kind == CompletionItemKind::Missing) |
| 391 | Completion.Kind = toCompletionItemKind(Kind: C.IndexResult->SymInfo.Kind, |
| 392 | Signature: &C.IndexResult->Signature); |
| 393 | if (Completion.Name.empty()) |
| 394 | Completion.Name = std::string(C.IndexResult->Name); |
| 395 | if (Completion.FilterText.empty()) |
| 396 | Completion.FilterText = Completion.Name; |
| 397 | // If the completion was visible to Sema, no qualifier is needed. This |
| 398 | // avoids unneeded qualifiers in cases like with `using ns::X`. |
| 399 | if (Completion.RequiredQualifier.empty() && !C.SemaResult) { |
| 400 | llvm::StringRef ShortestQualifier = C.IndexResult->Scope; |
| 401 | for (llvm::StringRef Scope : AccessibleScopes) { |
| 402 | llvm::StringRef Qualifier = C.IndexResult->Scope; |
| 403 | if (Qualifier.consume_front(Prefix: Scope) && |
| 404 | Qualifier.size() < ShortestQualifier.size()) |
| 405 | ShortestQualifier = Qualifier; |
| 406 | } |
| 407 | Completion.RequiredQualifier = std::string(ShortestQualifier); |
| 408 | } |
| 409 | } |
| 410 | if (C.IdentifierResult) { |
| 411 | Completion.Origin |= SymbolOrigin::Identifier; |
| 412 | Completion.Kind = CompletionItemKind::Text; |
| 413 | Completion.Name = std::string(C.IdentifierResult->Name); |
| 414 | Completion.FilterText = Completion.Name; |
| 415 | } |
| 416 | |
| 417 | // Turn absolute path into a literal string that can be #included. |
| 418 | auto Inserted = [&](llvm::StringRef ) |
| 419 | -> llvm::Expected<std::pair<std::string, bool>> { |
| 420 | auto ResolvedDeclaring = |
| 421 | URI::resolve(FileURI: C.IndexResult->CanonicalDeclaration.FileURI, HintPath: FileName); |
| 422 | if (!ResolvedDeclaring) |
| 423 | return ResolvedDeclaring.takeError(); |
| 424 | auto ResolvedInserted = toHeaderFile(Header, HintPath: FileName); |
| 425 | if (!ResolvedInserted) |
| 426 | return ResolvedInserted.takeError(); |
| 427 | auto Spelled = Includes.calculateIncludePath(InsertedHeader: *ResolvedInserted, IncludingFile: FileName); |
| 428 | if (!Spelled) |
| 429 | return error(Fmt: "Header not on include path" ); |
| 430 | return std::make_pair( |
| 431 | x: std::move(*Spelled), |
| 432 | y: Includes.shouldInsertInclude(DeclaringHeader: *ResolvedDeclaring, InsertedHeader: *ResolvedInserted)); |
| 433 | }; |
| 434 | bool ShouldInsert = |
| 435 | C.headerToInsertIfAllowed(Opts, ContextKind).has_value(); |
| 436 | Symbol::IncludeDirective Directive = insertionDirective(Opts); |
| 437 | // Calculate include paths and edits for all possible headers. |
| 438 | for (const auto &Inc : C.RankedIncludeHeaders) { |
| 439 | if ((Inc.Directive & Directive) == 0) |
| 440 | continue; |
| 441 | |
| 442 | if (auto ToInclude = Inserted(Inc.Header)) { |
| 443 | CodeCompletion::IncludeCandidate Include; |
| 444 | Include.Header = ToInclude->first; |
| 445 | if (ToInclude->second && ShouldInsert) |
| 446 | Include.Insertion = Includes.insert( |
| 447 | VerbatimHeader: ToInclude->first, Directive: Directive == Symbol::Import |
| 448 | ? tooling::IncludeDirective::Import |
| 449 | : tooling::IncludeDirective::Include); |
| 450 | Completion.Includes.push_back(Elt: std::move(Include)); |
| 451 | } else |
| 452 | log(Fmt: "Failed to generate include insertion edits for adding header " |
| 453 | "(FileURI='{0}', IncludeHeader='{1}') into {2}: {3}" , |
| 454 | Vals: C.IndexResult->CanonicalDeclaration.FileURI, Vals: Inc.Header, Vals&: FileName, |
| 455 | Vals: ToInclude.takeError()); |
| 456 | } |
| 457 | // Prefer includes that do not need edits (i.e. already exist). |
| 458 | std::stable_partition(first: Completion.Includes.begin(), |
| 459 | last: Completion.Includes.end(), |
| 460 | pred: [](const CodeCompletion::IncludeCandidate &I) { |
| 461 | return !I.Insertion.has_value(); |
| 462 | }); |
| 463 | } |
| 464 | |
| 465 | void add(const CompletionCandidate &C, CodeCompletionString *SemaCCS, |
| 466 | CodeCompletionContext::Kind ContextKind) { |
| 467 | assert(bool(C.SemaResult) == bool(SemaCCS)); |
| 468 | Bundled.emplace_back(); |
| 469 | BundledEntry &S = Bundled.back(); |
| 470 | bool IsConcept = false; |
| 471 | if (C.SemaResult) { |
| 472 | getSignature(CCS: *SemaCCS, Signature: &S.Signature, Snippet: &S.SnippetSuffix, ResultKind: C.SemaResult->Kind, |
| 473 | CursorKind: C.SemaResult->CursorKind, |
| 474 | /*IncludeFunctionArguments=*/C.SemaResult->FunctionCanBeCall, |
| 475 | /*RequiredQualifiers=*/&Completion.RequiredQualifier); |
| 476 | S.ReturnType = getReturnType(CCS: *SemaCCS); |
| 477 | if (C.SemaResult->Kind == CodeCompletionResult::RK_Declaration) |
| 478 | if (const auto *D = C.SemaResult->getDeclaration()) |
| 479 | if (isa<ConceptDecl>(Val: D)) |
| 480 | IsConcept = true; |
| 481 | } else if (C.IndexResult) { |
| 482 | S.Signature = std::string(C.IndexResult->Signature); |
| 483 | S.SnippetSuffix = std::string(C.IndexResult->CompletionSnippetSuffix); |
| 484 | S.ReturnType = std::string(C.IndexResult->ReturnType); |
| 485 | if (C.IndexResult->SymInfo.Kind == index::SymbolKind::Concept) |
| 486 | IsConcept = true; |
| 487 | } |
| 488 | |
| 489 | /// When a concept is used as a type-constraint (e.g. `Iterator auto x`), |
| 490 | /// and in some other contexts, its first type argument is not written. |
| 491 | /// Drop the parameter from the signature. |
| 492 | if (IsConcept && ContextKind == CodeCompletionContext::CCC_TopLevel) { |
| 493 | S.Signature = removeFirstTemplateArg(Signature: S.Signature); |
| 494 | // Dropping the first placeholder from the suffix will leave a $2 |
| 495 | // with no $1. |
| 496 | S.SnippetSuffix = removeFirstTemplateArg(Signature: S.SnippetSuffix); |
| 497 | } |
| 498 | |
| 499 | if (!Completion.Documentation) { |
| 500 | auto SetDoc = [&](llvm::StringRef Doc) { |
| 501 | if (!Doc.empty()) { |
| 502 | Completion.Documentation.emplace(); |
| 503 | parseDocumentation(Input: Doc, Output&: *Completion.Documentation); |
| 504 | } |
| 505 | }; |
| 506 | if (C.IndexResult) { |
| 507 | SetDoc(C.IndexResult->Documentation); |
| 508 | } else if (C.SemaResult) { |
| 509 | const auto = getDocComment(Ctx: *ASTCtx, Result: *C.SemaResult, |
| 510 | /*CommentsFromHeaders=*/false); |
| 511 | SetDoc(formatDocumentation(CCS: *SemaCCS, DocComment)); |
| 512 | } |
| 513 | } |
| 514 | if (Completion.Deprecated) { |
| 515 | if (C.SemaResult) |
| 516 | Completion.Deprecated &= |
| 517 | C.SemaResult->Availability == CXAvailability_Deprecated; |
| 518 | if (C.IndexResult) |
| 519 | Completion.Deprecated &= |
| 520 | bool(C.IndexResult->Flags & Symbol::Deprecated); |
| 521 | } |
| 522 | } |
| 523 | |
| 524 | CodeCompletion build() { |
| 525 | Completion.ReturnType = summarizeReturnType(); |
| 526 | Completion.Signature = summarizeSignature(); |
| 527 | Completion.SnippetSuffix = summarizeSnippet(); |
| 528 | Completion.BundleSize = Bundled.size(); |
| 529 | return std::move(Completion); |
| 530 | } |
| 531 | |
| 532 | private: |
| 533 | struct BundledEntry { |
| 534 | std::string SnippetSuffix; |
| 535 | std::string Signature; |
| 536 | std::string ReturnType; |
| 537 | }; |
| 538 | |
| 539 | // If all BundledEntries have the same value for a property, return it. |
| 540 | template <std::string BundledEntry::*Member> |
| 541 | const std::string *onlyValue() const { |
| 542 | auto B = Bundled.begin(), E = Bundled.end(); |
| 543 | for (auto *I = B + 1; I != E; ++I) |
| 544 | if (I->*Member != B->*Member) |
| 545 | return nullptr; |
| 546 | return &(B->*Member); |
| 547 | } |
| 548 | |
| 549 | template <bool BundledEntry::*Member> const bool *onlyValue() const { |
| 550 | auto B = Bundled.begin(), E = Bundled.end(); |
| 551 | for (auto *I = B + 1; I != E; ++I) |
| 552 | if (I->*Member != B->*Member) |
| 553 | return nullptr; |
| 554 | return &(B->*Member); |
| 555 | } |
| 556 | |
| 557 | std::string summarizeReturnType() const { |
| 558 | if (auto *RT = onlyValue<&BundledEntry::ReturnType>()) |
| 559 | return *RT; |
| 560 | return "" ; |
| 561 | } |
| 562 | |
| 563 | std::string summarizeSnippet() const { |
| 564 | /// localize ArgumentLists tests for better readability |
| 565 | const bool None = ArgumentLists == Config::ArgumentListsPolicy::None; |
| 566 | const bool Open = |
| 567 | ArgumentLists == Config::ArgumentListsPolicy::OpenDelimiter; |
| 568 | const bool Delim = ArgumentLists == Config::ArgumentListsPolicy::Delimiters; |
| 569 | const bool Full = |
| 570 | ArgumentLists == Config::ArgumentListsPolicy::FullPlaceholders || |
| 571 | (!None && !Open && !Delim); // <-- failsafe: Full is default |
| 572 | |
| 573 | if (IsUsingDeclaration) |
| 574 | return "" ; |
| 575 | auto *Snippet = onlyValue<&BundledEntry::SnippetSuffix>(); |
| 576 | if (!Snippet) |
| 577 | // All bundles are function calls. |
| 578 | // FIXME(ibiryukov): sometimes add template arguments to a snippet, e.g. |
| 579 | // we need to complete 'forward<$1>($0)'. |
| 580 | return None ? "" : (Open ? "(" : "($0)" ); |
| 581 | |
| 582 | if (Snippet->empty()) |
| 583 | return "" ; |
| 584 | |
| 585 | bool MayHaveArgList = Completion.Kind == CompletionItemKind::Function || |
| 586 | Completion.Kind == CompletionItemKind::Method || |
| 587 | Completion.Kind == CompletionItemKind::Constructor || |
| 588 | Completion.Kind == CompletionItemKind::Text /*Macro*/; |
| 589 | // If likely arg list already exists, don't add new parens & placeholders. |
| 590 | // Snippet: function(int x, int y) |
| 591 | // func^(1,2) -> function(1, 2) |
| 592 | // NOT function(int x, int y)(1, 2) |
| 593 | if (MayHaveArgList) { |
| 594 | // Check for a template argument list in the code. |
| 595 | // Snippet: function<class T>(int x) |
| 596 | // fu^<int>(1) -> function<int>(1) |
| 597 | if (NextTokenKind == tok::less && Snippet->front() == '<') |
| 598 | return "" ; |
| 599 | // Potentially followed by regular argument list. |
| 600 | if (NextTokenKind == tok::l_paren) { |
| 601 | // Snippet: function<class T>(int x) |
| 602 | // fu^(1,2) -> function<class T>(1, 2) |
| 603 | if (Snippet->front() == '<') { |
| 604 | // Find matching '>', handling nested brackets. |
| 605 | int Balance = 0; |
| 606 | size_t I = 0; |
| 607 | do { |
| 608 | if (Snippet->at(n: I) == '>') |
| 609 | --Balance; |
| 610 | else if (Snippet->at(n: I) == '<') |
| 611 | ++Balance; |
| 612 | ++I; |
| 613 | } while (Balance > 0); |
| 614 | return Snippet->substr(pos: 0, n: I); |
| 615 | } |
| 616 | return "" ; |
| 617 | } |
| 618 | } |
| 619 | if (Full) |
| 620 | return *Snippet; |
| 621 | |
| 622 | // Replace argument snippets with a simplified pattern. |
| 623 | if (MayHaveArgList) { |
| 624 | // Functions snippets can be of 2 types: |
| 625 | // - containing only function arguments, e.g. |
| 626 | // foo(${1:int p1}, ${2:int p2}); |
| 627 | // We transform this pattern to '($0)' or '()'. |
| 628 | // - template arguments and function arguments, e.g. |
| 629 | // foo<${1:class}>(${2:int p1}). |
| 630 | // We transform this pattern to '<$1>()$0' or '<$0>()'. |
| 631 | |
| 632 | bool EmptyArgs = llvm::StringRef(*Snippet).ends_with(Suffix: "()" ); |
| 633 | if (Snippet->front() == '<') |
| 634 | return None ? "" : (Open ? "<" : (EmptyArgs ? "<$1>()$0" : "<$1>($0)" )); |
| 635 | if (Snippet->front() == '(') |
| 636 | return None ? "" : (Open ? "(" : (EmptyArgs ? "()" : "($0)" )); |
| 637 | return *Snippet; // Not an arg snippet? |
| 638 | } |
| 639 | // 'CompletionItemKind::Interface' matches template type aliases. |
| 640 | if (Completion.Kind == CompletionItemKind::Interface || |
| 641 | Completion.Kind == CompletionItemKind::Class || |
| 642 | Completion.Kind == CompletionItemKind::Variable) { |
| 643 | if (Snippet->front() != '<') |
| 644 | return *Snippet; // Not an arg snippet? |
| 645 | |
| 646 | // Classes and template using aliases can only have template arguments, |
| 647 | // e.g. Foo<${1:class}>. |
| 648 | if (llvm::StringRef(*Snippet).ends_with(Suffix: "<>" )) |
| 649 | return "<>" ; // can happen with defaulted template arguments. |
| 650 | return None ? "" : (Open ? "<" : "<$0>" ); |
| 651 | } |
| 652 | return *Snippet; |
| 653 | } |
| 654 | |
| 655 | std::string summarizeSignature() const { |
| 656 | if (auto *Signature = onlyValue<&BundledEntry::Signature>()) |
| 657 | return *Signature; |
| 658 | // All bundles are function calls. |
| 659 | return "(…)" ; |
| 660 | } |
| 661 | |
| 662 | // ASTCtx can be nullptr if not run with sema. |
| 663 | ASTContext *ASTCtx; |
| 664 | CodeCompletion Completion; |
| 665 | llvm::SmallVector<BundledEntry, 1> Bundled; |
| 666 | /// the way argument lists are handled. |
| 667 | Config::ArgumentListsPolicy ArgumentLists; |
| 668 | // No snippets will be generated for using declarations and when the function |
| 669 | // arguments are already present. |
| 670 | bool IsUsingDeclaration; |
| 671 | tok::TokenKind NextTokenKind; |
| 672 | }; |
| 673 | |
| 674 | // Determine the symbol ID for a Sema code completion result, if possible. |
| 675 | SymbolID getSymbolID(const CodeCompletionResult &R, const SourceManager &SM) { |
| 676 | switch (R.Kind) { |
| 677 | case CodeCompletionResult::RK_Declaration: |
| 678 | case CodeCompletionResult::RK_Pattern: { |
| 679 | // Computing USR caches linkage, which may change after code completion. |
| 680 | if (hasUnstableLinkage(R.Declaration)) |
| 681 | return {}; |
| 682 | return clang::clangd::getSymbolID(R.Declaration); |
| 683 | } |
| 684 | case CodeCompletionResult::RK_Macro: |
| 685 | return clang::clangd::getSymbolID(MacroName: R.Macro->getName(), MI: R.MacroDefInfo, SM); |
| 686 | case CodeCompletionResult::RK_Keyword: |
| 687 | return {}; |
| 688 | } |
| 689 | llvm_unreachable("unknown CodeCompletionResult kind" ); |
| 690 | } |
| 691 | |
| 692 | // Scopes of the partial identifier we're trying to complete. |
| 693 | // It is used when we query the index for more completion results. |
| 694 | struct SpecifiedScope { |
| 695 | // The scopes we should look in, determined by Sema. |
| 696 | // |
| 697 | // If the qualifier was fully resolved, we look for completions in these |
| 698 | // scopes; if there is an unresolved part of the qualifier, it should be |
| 699 | // resolved within these scopes. |
| 700 | // |
| 701 | // Examples of qualified completion: |
| 702 | // |
| 703 | // "::vec" => {""} |
| 704 | // "using namespace std; ::vec^" => {"", "std::"} |
| 705 | // "namespace ns {using namespace std;} ns::^" => {"ns::", "std::"} |
| 706 | // "std::vec^" => {""} // "std" unresolved |
| 707 | // |
| 708 | // Examples of unqualified completion: |
| 709 | // |
| 710 | // "vec^" => {""} |
| 711 | // "using namespace std; vec^" => {"", "std::"} |
| 712 | // "namespace ns {inline namespace ni { struct Foo {}}} |
| 713 | // using namespace ns::ni; Fo^ " => {"", "ns::ni::"} |
| 714 | // "using namespace std; namespace ns { vec^ }" => {"ns::", "std::", ""} |
| 715 | // |
| 716 | // "" for global namespace, "ns::" for normal namespace. |
| 717 | std::vector<std::string> AccessibleScopes; |
| 718 | // This is an overestimate of AccessibleScopes, e.g. it ignores inline |
| 719 | // namespaces, to fetch more relevant symbols from index. |
| 720 | std::vector<std::string> QueryScopes; |
| 721 | // The full scope qualifier as typed by the user (without the leading "::"). |
| 722 | // Set if the qualifier is not fully resolved by Sema. |
| 723 | std::optional<std::string> UnresolvedQualifier; |
| 724 | |
| 725 | std::optional<std::string> EnclosingNamespace; |
| 726 | |
| 727 | bool AllowAllScopes = false; |
| 728 | |
| 729 | // Scopes that are accessible from current context. Used for dropping |
| 730 | // unnecessary namespecifiers. |
| 731 | std::vector<std::string> scopesForQualification() { |
| 732 | std::set<std::string> Results; |
| 733 | for (llvm::StringRef AS : AccessibleScopes) |
| 734 | Results.insert( |
| 735 | x: (AS + (UnresolvedQualifier ? *UnresolvedQualifier : "" )).str()); |
| 736 | return {Results.begin(), Results.end()}; |
| 737 | } |
| 738 | |
| 739 | // Construct scopes being queried in indexes. The results are deduplicated. |
| 740 | // This method formats the scopes to match the index request representation. |
| 741 | std::vector<std::string> scopesForIndexQuery() { |
| 742 | // The enclosing namespace must be first, it gets a quality boost. |
| 743 | std::vector<std::string> EnclosingAtFront; |
| 744 | if (EnclosingNamespace.has_value()) |
| 745 | EnclosingAtFront.push_back(x: *EnclosingNamespace); |
| 746 | std::set<std::string> Deduplicated; |
| 747 | for (llvm::StringRef S : QueryScopes) |
| 748 | if (S != EnclosingNamespace) |
| 749 | Deduplicated.insert(x: (S + UnresolvedQualifier.value_or(u: "" )).str()); |
| 750 | |
| 751 | EnclosingAtFront.reserve(n: EnclosingAtFront.size() + Deduplicated.size()); |
| 752 | llvm::copy(Range&: Deduplicated, Out: std::back_inserter(x&: EnclosingAtFront)); |
| 753 | |
| 754 | return EnclosingAtFront; |
| 755 | } |
| 756 | }; |
| 757 | |
| 758 | // Get all scopes that will be queried in indexes and whether symbols from |
| 759 | // any scope is allowed. The first scope in the list is the preferred scope |
| 760 | // (e.g. enclosing namespace). |
| 761 | SpecifiedScope getQueryScopes(CodeCompletionContext &CCContext, |
| 762 | const Sema &CCSema, |
| 763 | const CompletionPrefix &HeuristicPrefix, |
| 764 | const CodeCompleteOptions &Opts) { |
| 765 | SpecifiedScope Scopes; |
| 766 | for (auto *Context : CCContext.getVisitedContexts()) { |
| 767 | if (isa<TranslationUnitDecl>(Val: Context)) { |
| 768 | Scopes.QueryScopes.push_back(x: "" ); |
| 769 | Scopes.AccessibleScopes.push_back(x: "" ); |
| 770 | } else if (const auto *ND = dyn_cast<NamespaceDecl>(Val: Context)) { |
| 771 | Scopes.QueryScopes.push_back(x: printNamespaceScope(DC: *Context)); |
| 772 | Scopes.AccessibleScopes.push_back(x: printQualifiedName(*ND) + "::" ); |
| 773 | } |
| 774 | } |
| 775 | |
| 776 | const CXXScopeSpec *SemaSpecifier = |
| 777 | CCContext.getCXXScopeSpecifier().value_or(u: nullptr); |
| 778 | // Case 1: unqualified completion. |
| 779 | if (!SemaSpecifier) { |
| 780 | // Case 2 (exception): sema saw no qualifier, but there appears to be one! |
| 781 | // This can happen e.g. in incomplete macro expansions. Use heuristics. |
| 782 | if (!HeuristicPrefix.Qualifier.empty()) { |
| 783 | vlog(Fmt: "Sema said no scope specifier, but we saw {0} in the source code" , |
| 784 | Vals: HeuristicPrefix.Qualifier); |
| 785 | StringRef SpelledSpecifier = HeuristicPrefix.Qualifier; |
| 786 | if (SpelledSpecifier.consume_front(Prefix: "::" )) { |
| 787 | Scopes.AccessibleScopes = {"" }; |
| 788 | Scopes.QueryScopes = {"" }; |
| 789 | } |
| 790 | Scopes.UnresolvedQualifier = std::string(SpelledSpecifier); |
| 791 | return Scopes; |
| 792 | } |
| 793 | /// FIXME: When the enclosing namespace contains an inline namespace, |
| 794 | /// it's dropped here. This leads to a behavior similar to |
| 795 | /// https://github.com/clangd/clangd/issues/1451 |
| 796 | Scopes.EnclosingNamespace = printNamespaceScope(DC: *CCSema.CurContext); |
| 797 | // Allow AllScopes completion as there is no explicit scope qualifier. |
| 798 | Scopes.AllowAllScopes = Opts.AllScopes; |
| 799 | return Scopes; |
| 800 | } |
| 801 | // Case 3: sema saw and resolved a scope qualifier. |
| 802 | if (SemaSpecifier && SemaSpecifier->isValid()) |
| 803 | return Scopes; |
| 804 | |
| 805 | // Case 4: There was a qualifier, and Sema didn't resolve it. |
| 806 | Scopes.QueryScopes.push_back(x: "" ); // Make sure global scope is included. |
| 807 | llvm::StringRef SpelledSpecifier = Lexer::getSourceText( |
| 808 | Range: CharSourceRange::getCharRange(R: SemaSpecifier->getRange()), |
| 809 | SM: CCSema.SourceMgr, LangOpts: clang::LangOptions()); |
| 810 | if (SpelledSpecifier.consume_front(Prefix: "::" )) |
| 811 | Scopes.QueryScopes = {"" }; |
| 812 | Scopes.UnresolvedQualifier = std::string(SpelledSpecifier); |
| 813 | // Sema excludes the trailing "::". |
| 814 | if (!Scopes.UnresolvedQualifier->empty()) |
| 815 | *Scopes.UnresolvedQualifier += "::" ; |
| 816 | |
| 817 | Scopes.AccessibleScopes = Scopes.QueryScopes; |
| 818 | |
| 819 | return Scopes; |
| 820 | } |
| 821 | |
| 822 | // Should we perform index-based completion in a context of the specified kind? |
| 823 | // FIXME: consider allowing completion, but restricting the result types. |
| 824 | bool contextAllowsIndex(enum CodeCompletionContext::Kind K) { |
| 825 | switch (K) { |
| 826 | case CodeCompletionContext::CCC_TopLevel: |
| 827 | case CodeCompletionContext::CCC_ObjCInterface: |
| 828 | case CodeCompletionContext::CCC_ObjCImplementation: |
| 829 | case CodeCompletionContext::CCC_ObjCIvarList: |
| 830 | case CodeCompletionContext::CCC_ClassStructUnion: |
| 831 | case CodeCompletionContext::CCC_Statement: |
| 832 | case CodeCompletionContext::CCC_Expression: |
| 833 | case CodeCompletionContext::CCC_ObjCMessageReceiver: |
| 834 | case CodeCompletionContext::CCC_EnumTag: |
| 835 | case CodeCompletionContext::CCC_UnionTag: |
| 836 | case CodeCompletionContext::CCC_ClassOrStructTag: |
| 837 | case CodeCompletionContext::CCC_ObjCProtocolName: |
| 838 | case CodeCompletionContext::CCC_Namespace: |
| 839 | case CodeCompletionContext::CCC_Type: |
| 840 | case CodeCompletionContext::CCC_ParenthesizedExpression: |
| 841 | case CodeCompletionContext::CCC_ObjCInterfaceName: |
| 842 | case CodeCompletionContext::CCC_Symbol: |
| 843 | case CodeCompletionContext::CCC_SymbolOrNewName: |
| 844 | case CodeCompletionContext::CCC_ObjCClassForwardDecl: |
| 845 | case CodeCompletionContext::CCC_TopLevelOrExpression: |
| 846 | return true; |
| 847 | case CodeCompletionContext::CCC_OtherWithMacros: |
| 848 | case CodeCompletionContext::CCC_DotMemberAccess: |
| 849 | case CodeCompletionContext::CCC_ArrowMemberAccess: |
| 850 | case CodeCompletionContext::CCC_ObjCCategoryName: |
| 851 | case CodeCompletionContext::CCC_ObjCPropertyAccess: |
| 852 | case CodeCompletionContext::CCC_MacroName: |
| 853 | case CodeCompletionContext::CCC_MacroNameUse: |
| 854 | case CodeCompletionContext::CCC_PreprocessorExpression: |
| 855 | case CodeCompletionContext::CCC_PreprocessorDirective: |
| 856 | case CodeCompletionContext::CCC_SelectorName: |
| 857 | case CodeCompletionContext::CCC_TypeQualifiers: |
| 858 | case CodeCompletionContext::CCC_ObjCInstanceMessage: |
| 859 | case CodeCompletionContext::CCC_ObjCClassMessage: |
| 860 | case CodeCompletionContext::CCC_IncludedFile: |
| 861 | case CodeCompletionContext::CCC_Attribute: |
| 862 | // FIXME: Provide identifier based completions for the following contexts: |
| 863 | case CodeCompletionContext::CCC_Other: // Be conservative. |
| 864 | case CodeCompletionContext::CCC_NaturalLanguage: |
| 865 | case CodeCompletionContext::CCC_Recovery: |
| 866 | case CodeCompletionContext::CCC_NewName: |
| 867 | return false; |
| 868 | } |
| 869 | llvm_unreachable("unknown code completion context" ); |
| 870 | } |
| 871 | |
| 872 | static bool isInjectedClass(const NamedDecl &D) { |
| 873 | if (auto *R = dyn_cast_or_null<RecordDecl>(Val: &D)) |
| 874 | if (R->isInjectedClassName()) |
| 875 | return true; |
| 876 | return false; |
| 877 | } |
| 878 | |
| 879 | // Some member calls are excluded because they're so rarely useful. |
| 880 | static bool isExcludedMember(const NamedDecl &D) { |
| 881 | // Destructor completion is rarely useful, and works inconsistently. |
| 882 | // (s.^ completes ~string, but s.~st^ is an error). |
| 883 | if (D.getKind() == Decl::CXXDestructor) |
| 884 | return true; |
| 885 | // Injected name may be useful for A::foo(), but who writes A::A::foo()? |
| 886 | if (isInjectedClass(D)) |
| 887 | return true; |
| 888 | // Explicit calls to operators are also rare. |
| 889 | auto NameKind = D.getDeclName().getNameKind(); |
| 890 | if (NameKind == DeclarationName::CXXOperatorName || |
| 891 | NameKind == DeclarationName::CXXLiteralOperatorName || |
| 892 | NameKind == DeclarationName::CXXConversionFunctionName) |
| 893 | return true; |
| 894 | return false; |
| 895 | } |
| 896 | |
| 897 | // The CompletionRecorder captures Sema code-complete output, including context. |
| 898 | // It filters out ignored results (but doesn't apply fuzzy-filtering yet). |
| 899 | // It doesn't do scoring or conversion to CompletionItem yet, as we want to |
| 900 | // merge with index results first. |
| 901 | // Generally the fields and methods of this object should only be used from |
| 902 | // within the callback. |
| 903 | struct CompletionRecorder : public CodeCompleteConsumer { |
| 904 | CompletionRecorder(const CodeCompleteOptions &Opts, |
| 905 | llvm::unique_function<void()> ResultsCallback) |
| 906 | : CodeCompleteConsumer(Opts.getClangCompleteOpts()), |
| 907 | CCContext(CodeCompletionContext::CCC_Other), Opts(Opts), |
| 908 | CCAllocator(std::make_shared<GlobalCodeCompletionAllocator>()), |
| 909 | CCTUInfo(CCAllocator), ResultsCallback(std::move(ResultsCallback)) { |
| 910 | assert(this->ResultsCallback); |
| 911 | } |
| 912 | |
| 913 | std::vector<CodeCompletionResult> Results; |
| 914 | CodeCompletionContext CCContext; |
| 915 | Sema *CCSema = nullptr; // Sema that created the results. |
| 916 | // FIXME: Sema is scary. Can we store ASTContext and Preprocessor, instead? |
| 917 | |
| 918 | void ProcessCodeCompleteResults(class Sema &S, CodeCompletionContext Context, |
| 919 | CodeCompletionResult *InResults, |
| 920 | unsigned NumResults) final { |
| 921 | // Results from recovery mode are generally useless, and the callback after |
| 922 | // recovery (if any) is usually more interesting. To make sure we handle the |
| 923 | // future callback from sema, we just ignore all callbacks in recovery mode, |
| 924 | // as taking only results from recovery mode results in poor completion |
| 925 | // results. |
| 926 | // FIXME: in case there is no future sema completion callback after the |
| 927 | // recovery mode, we might still want to provide some results (e.g. trivial |
| 928 | // identifier-based completion). |
| 929 | CodeCompletionContext::Kind ContextKind = Context.getKind(); |
| 930 | if (ContextKind == CodeCompletionContext::CCC_Recovery) { |
| 931 | log(Fmt: "Code complete: Ignoring sema code complete callback with Recovery " |
| 932 | "context." ); |
| 933 | return; |
| 934 | } |
| 935 | // If a callback is called without any sema result and the context does not |
| 936 | // support index-based completion, we simply skip it to give way to |
| 937 | // potential future callbacks with results. |
| 938 | if (NumResults == 0 && !contextAllowsIndex(K: Context.getKind())) |
| 939 | return; |
| 940 | if (CCSema) { |
| 941 | log("Multiple code complete callbacks (parser backtracked?). " |
| 942 | "Dropping results from context {0}, keeping results from {1}." , |
| 943 | getCompletionKindString(Kind: Context.getKind()), |
| 944 | getCompletionKindString(this->CCContext.getKind())); |
| 945 | return; |
| 946 | } |
| 947 | // Record the completion context. |
| 948 | CCSema = &S; |
| 949 | CCContext = Context; |
| 950 | |
| 951 | // Retain the results we might want. |
| 952 | for (unsigned I = 0; I < NumResults; ++I) { |
| 953 | auto &Result = InResults[I]; |
| 954 | if (Config::current().Completion.CodePatterns == |
| 955 | Config::CodePatternsPolicy::None && |
| 956 | Result.Kind == CodeCompletionResult::RK_Pattern && |
| 957 | // keep allowing the include files autocomplete suggestions |
| 958 | ContextKind != CodeCompletionContext::CCC_IncludedFile) |
| 959 | continue; |
| 960 | // Class members that are shadowed by subclasses are usually noise. |
| 961 | if (Result.Hidden && Result.Declaration && |
| 962 | Result.Declaration->isCXXClassMember()) |
| 963 | continue; |
| 964 | if (!Opts.IncludeIneligibleResults && |
| 965 | (Result.Availability == CXAvailability_NotAvailable || |
| 966 | Result.Availability == CXAvailability_NotAccessible)) |
| 967 | continue; |
| 968 | if (Result.Declaration && |
| 969 | !Context.getBaseType().isNull() // is this a member-access context? |
| 970 | && isExcludedMember(D: *Result.Declaration)) |
| 971 | continue; |
| 972 | // Skip injected class name when no class scope is not explicitly set. |
| 973 | // E.g. show injected A::A in `using A::A^` but not in "A^". |
| 974 | if (Result.Declaration && !Context.getCXXScopeSpecifier() && |
| 975 | isInjectedClass(D: *Result.Declaration)) |
| 976 | continue; |
| 977 | // We choose to never append '::' to completion results in clangd. |
| 978 | Result.StartsNestedNameSpecifier = false; |
| 979 | Results.push_back(x: Result); |
| 980 | } |
| 981 | ResultsCallback(); |
| 982 | } |
| 983 | |
| 984 | CodeCompletionAllocator &getAllocator() override { return *CCAllocator; } |
| 985 | CodeCompletionTUInfo &getCodeCompletionTUInfo() override { return CCTUInfo; } |
| 986 | |
| 987 | // Returns the filtering/sorting name for Result, which must be from Results. |
| 988 | // Returned string is owned by this recorder (or the AST). |
| 989 | llvm::StringRef getName(const CodeCompletionResult &Result) { |
| 990 | switch (Result.Kind) { |
| 991 | case CodeCompletionResult::RK_Declaration: |
| 992 | if (auto *ID = Result.Declaration->getIdentifier()) |
| 993 | return ID->getName(); |
| 994 | break; |
| 995 | case CodeCompletionResult::RK_Keyword: |
| 996 | return Result.Keyword; |
| 997 | case CodeCompletionResult::RK_Macro: |
| 998 | return Result.Macro->getName(); |
| 999 | case CodeCompletionResult::RK_Pattern: |
| 1000 | break; |
| 1001 | } |
| 1002 | auto *CCS = codeCompletionString(R: Result); |
| 1003 | const CodeCompletionString::Chunk *OnlyText = nullptr; |
| 1004 | for (auto &C : *CCS) { |
| 1005 | if (C.Kind != CodeCompletionString::CK_TypedText) |
| 1006 | continue; |
| 1007 | if (OnlyText) |
| 1008 | return CCAllocator->CopyString(String: CCS->getAllTypedText()); |
| 1009 | OnlyText = &C; |
| 1010 | } |
| 1011 | return OnlyText ? OnlyText->Text : llvm::StringRef(); |
| 1012 | } |
| 1013 | |
| 1014 | // Build a CodeCompletion string for R, which must be from Results. |
| 1015 | // The CCS will be owned by this recorder. |
| 1016 | CodeCompletionString *codeCompletionString(const CodeCompletionResult &R) { |
| 1017 | // CodeCompletionResult doesn't seem to be const-correct. We own it, anyway. |
| 1018 | return const_cast<CodeCompletionResult &>(R).CreateCodeCompletionString( |
| 1019 | *CCSema, CCContext, *CCAllocator, CCTUInfo, |
| 1020 | /*IncludeBriefComments=*/false); |
| 1021 | } |
| 1022 | |
| 1023 | private: |
| 1024 | CodeCompleteOptions Opts; |
| 1025 | std::shared_ptr<GlobalCodeCompletionAllocator> CCAllocator; |
| 1026 | CodeCompletionTUInfo CCTUInfo; |
| 1027 | llvm::unique_function<void()> ResultsCallback; |
| 1028 | }; |
| 1029 | |
| 1030 | struct ScoredSignature { |
| 1031 | // When not null, requires documentation to be requested from the index with |
| 1032 | // this ID. |
| 1033 | SymbolID IDForDoc; |
| 1034 | SignatureInformation Signature; |
| 1035 | SignatureQualitySignals Quality; |
| 1036 | }; |
| 1037 | |
| 1038 | // Returns the index of the parameter matching argument number "Arg. |
| 1039 | // This is usually just "Arg", except for variadic functions/templates, where |
| 1040 | // "Arg" might be higher than the number of parameters. When that happens, we |
| 1041 | // assume the last parameter is variadic and assume all further args are |
| 1042 | // part of it. |
| 1043 | int paramIndexForArg(const CodeCompleteConsumer::OverloadCandidate &Candidate, |
| 1044 | int Arg) { |
| 1045 | int NumParams = Candidate.getNumParams(); |
| 1046 | if (auto *T = Candidate.getFunctionType()) { |
| 1047 | if (auto *Proto = T->getAs<FunctionProtoType>()) { |
| 1048 | if (Proto->isVariadic()) |
| 1049 | ++NumParams; |
| 1050 | } |
| 1051 | } |
| 1052 | return std::min(a: Arg, b: std::max(a: NumParams - 1, b: 0)); |
| 1053 | } |
| 1054 | |
| 1055 | class SignatureHelpCollector final : public CodeCompleteConsumer { |
| 1056 | public: |
| 1057 | SignatureHelpCollector(const clang::CodeCompleteOptions &CodeCompleteOpts, |
| 1058 | MarkupKind DocumentationFormat, |
| 1059 | const SymbolIndex *Index, SignatureHelp &SigHelp) |
| 1060 | : CodeCompleteConsumer(CodeCompleteOpts), SigHelp(SigHelp), |
| 1061 | Allocator(std::make_shared<clang::GlobalCodeCompletionAllocator>()), |
| 1062 | CCTUInfo(Allocator), Index(Index), |
| 1063 | DocumentationFormat(DocumentationFormat) {} |
| 1064 | |
| 1065 | void ProcessOverloadCandidates(Sema &S, unsigned CurrentArg, |
| 1066 | OverloadCandidate *Candidates, |
| 1067 | unsigned NumCandidates, |
| 1068 | SourceLocation OpenParLoc, |
| 1069 | bool Braced) override { |
| 1070 | assert(!OpenParLoc.isInvalid()); |
| 1071 | SourceManager &SrcMgr = S.getSourceManager(); |
| 1072 | OpenParLoc = SrcMgr.getFileLoc(Loc: OpenParLoc); |
| 1073 | if (SrcMgr.isInMainFile(Loc: OpenParLoc)) |
| 1074 | SigHelp.argListStart = sourceLocToPosition(SM: SrcMgr, Loc: OpenParLoc); |
| 1075 | else |
| 1076 | elog(Fmt: "Location oustide main file in signature help: {0}" , |
| 1077 | Vals: OpenParLoc.printToString(SM: SrcMgr)); |
| 1078 | |
| 1079 | std::vector<ScoredSignature> ScoredSignatures; |
| 1080 | SigHelp.signatures.reserve(n: NumCandidates); |
| 1081 | ScoredSignatures.reserve(n: NumCandidates); |
| 1082 | // FIXME(rwols): How can we determine the "active overload candidate"? |
| 1083 | // Right now the overloaded candidates seem to be provided in a "best fit" |
| 1084 | // order, so I'm not too worried about this. |
| 1085 | SigHelp.activeSignature = 0; |
| 1086 | assert(CurrentArg <= (unsigned)std::numeric_limits<int>::max() && |
| 1087 | "too many arguments" ); |
| 1088 | |
| 1089 | SigHelp.activeParameter = static_cast<int>(CurrentArg); |
| 1090 | |
| 1091 | for (unsigned I = 0; I < NumCandidates; ++I) { |
| 1092 | OverloadCandidate Candidate = Candidates[I]; |
| 1093 | // We want to avoid showing instantiated signatures, because they may be |
| 1094 | // long in some cases (e.g. when 'T' is substituted with 'std::string', we |
| 1095 | // would get 'std::basic_string<char>'). |
| 1096 | if (auto *Func = Candidate.getFunction()) { |
| 1097 | if (auto *Pattern = Func->getTemplateInstantiationPattern()) |
| 1098 | Candidate = OverloadCandidate(Pattern); |
| 1099 | } |
| 1100 | if (static_cast<int>(I) == SigHelp.activeSignature) { |
| 1101 | // The activeParameter in LSP relates to the activeSignature. There is |
| 1102 | // another, per-signature field, but we currently do not use it and not |
| 1103 | // all clients might support it. |
| 1104 | // FIXME: Add support for per-signature activeParameter field. |
| 1105 | SigHelp.activeParameter = |
| 1106 | paramIndexForArg(Candidate, Arg: SigHelp.activeParameter); |
| 1107 | } |
| 1108 | |
| 1109 | const auto *CCS = Candidate.CreateSignatureString( |
| 1110 | CurrentArg, S, Allocator&: *Allocator, CCTUInfo, |
| 1111 | /*IncludeBriefComments=*/true, Braced); |
| 1112 | assert(CCS && "Expected the CodeCompletionString to be non-null" ); |
| 1113 | ScoredSignatures.push_back(x: processOverloadCandidate( |
| 1114 | Candidate, CCS: *CCS, |
| 1115 | DocComment: Candidate.getFunction() |
| 1116 | ? getDeclComment(S.getASTContext(), *Candidate.getFunction()) |
| 1117 | : "" )); |
| 1118 | } |
| 1119 | |
| 1120 | // Sema does not load the docs from the preamble, so we need to fetch extra |
| 1121 | // docs from the index instead. |
| 1122 | llvm::DenseMap<SymbolID, std::string> FetchedDocs; |
| 1123 | if (Index) { |
| 1124 | LookupRequest IndexRequest; |
| 1125 | for (const auto &S : ScoredSignatures) { |
| 1126 | if (!S.IDForDoc) |
| 1127 | continue; |
| 1128 | IndexRequest.IDs.insert(V: S.IDForDoc); |
| 1129 | } |
| 1130 | Index->lookup(Req: IndexRequest, Callback: [&](const Symbol &S) { |
| 1131 | if (!S.Documentation.empty()) |
| 1132 | FetchedDocs[S.ID] = std::string(S.Documentation); |
| 1133 | }); |
| 1134 | vlog(Fmt: "SigHelp: requested docs for {0} symbols from the index, got {1} " |
| 1135 | "symbols with non-empty docs in the response" , |
| 1136 | Vals: IndexRequest.IDs.size(), Vals: FetchedDocs.size()); |
| 1137 | } |
| 1138 | |
| 1139 | llvm::sort(C&: ScoredSignatures, Comp: [](const ScoredSignature &L, |
| 1140 | const ScoredSignature &R) { |
| 1141 | // Ordering follows: |
| 1142 | // - Less number of parameters is better. |
| 1143 | // - Aggregate > Function > FunctionType > FunctionTemplate |
| 1144 | // - High score is better. |
| 1145 | // - Shorter signature is better. |
| 1146 | // - Alphabetically smaller is better. |
| 1147 | if (L.Quality.NumberOfParameters != R.Quality.NumberOfParameters) |
| 1148 | return L.Quality.NumberOfParameters < R.Quality.NumberOfParameters; |
| 1149 | if (L.Quality.NumberOfOptionalParameters != |
| 1150 | R.Quality.NumberOfOptionalParameters) |
| 1151 | return L.Quality.NumberOfOptionalParameters < |
| 1152 | R.Quality.NumberOfOptionalParameters; |
| 1153 | if (L.Quality.Kind != R.Quality.Kind) { |
| 1154 | using OC = CodeCompleteConsumer::OverloadCandidate; |
| 1155 | auto KindPriority = [&](OC::CandidateKind K) { |
| 1156 | switch (K) { |
| 1157 | case OC::CK_Aggregate: |
| 1158 | return 0; |
| 1159 | case OC::CK_Function: |
| 1160 | return 1; |
| 1161 | case OC::CK_FunctionType: |
| 1162 | return 2; |
| 1163 | case OC::CK_FunctionProtoTypeLoc: |
| 1164 | return 3; |
| 1165 | case OC::CK_FunctionTemplate: |
| 1166 | return 4; |
| 1167 | case OC::CK_Template: |
| 1168 | return 5; |
| 1169 | } |
| 1170 | llvm_unreachable("Unknown overload candidate type." ); |
| 1171 | }; |
| 1172 | return KindPriority(L.Quality.Kind) < KindPriority(R.Quality.Kind); |
| 1173 | } |
| 1174 | if (L.Signature.label.size() != R.Signature.label.size()) |
| 1175 | return L.Signature.label.size() < R.Signature.label.size(); |
| 1176 | return L.Signature.label < R.Signature.label; |
| 1177 | }); |
| 1178 | |
| 1179 | for (auto &SS : ScoredSignatures) { |
| 1180 | auto IndexDocIt = |
| 1181 | SS.IDForDoc ? FetchedDocs.find(Val: SS.IDForDoc) : FetchedDocs.end(); |
| 1182 | if (IndexDocIt != FetchedDocs.end()) { |
| 1183 | markup::Document ; |
| 1184 | parseDocumentation(Input: IndexDocIt->second, Output&: SignatureComment); |
| 1185 | SS.Signature.documentation = |
| 1186 | renderDoc(Doc: SignatureComment, Kind: DocumentationFormat); |
| 1187 | } |
| 1188 | |
| 1189 | SigHelp.signatures.push_back(x: std::move(SS.Signature)); |
| 1190 | } |
| 1191 | } |
| 1192 | |
| 1193 | GlobalCodeCompletionAllocator &getAllocator() override { return *Allocator; } |
| 1194 | |
| 1195 | CodeCompletionTUInfo &getCodeCompletionTUInfo() override { return CCTUInfo; } |
| 1196 | |
| 1197 | private: |
| 1198 | void processParameterChunk(llvm::StringRef ChunkText, |
| 1199 | SignatureInformation &Signature) const { |
| 1200 | // (!) this is O(n), should still be fast compared to building ASTs. |
| 1201 | unsigned ParamStartOffset = lspLength(Code: Signature.label); |
| 1202 | unsigned ParamEndOffset = ParamStartOffset + lspLength(Code: ChunkText); |
| 1203 | // A piece of text that describes the parameter that corresponds to |
| 1204 | // the code-completion location within a function call, message send, |
| 1205 | // macro invocation, etc. |
| 1206 | Signature.label += ChunkText; |
| 1207 | ParameterInformation Info; |
| 1208 | Info.labelOffsets.emplace(args&: ParamStartOffset, args&: ParamEndOffset); |
| 1209 | // FIXME: only set 'labelOffsets' when all clients migrate out of it. |
| 1210 | Info.labelString = std::string(ChunkText); |
| 1211 | |
| 1212 | Signature.parameters.push_back(x: std::move(Info)); |
| 1213 | } |
| 1214 | |
| 1215 | void processOptionalChunk(const CodeCompletionString &CCS, |
| 1216 | SignatureInformation &Signature, |
| 1217 | SignatureQualitySignals &Signal) const { |
| 1218 | for (const auto &Chunk : CCS) { |
| 1219 | switch (Chunk.Kind) { |
| 1220 | case CodeCompletionString::CK_Optional: |
| 1221 | assert(Chunk.Optional && |
| 1222 | "Expected the optional code completion string to be non-null." ); |
| 1223 | processOptionalChunk(CCS: *Chunk.Optional, Signature, Signal); |
| 1224 | break; |
| 1225 | case CodeCompletionString::CK_VerticalSpace: |
| 1226 | break; |
| 1227 | case CodeCompletionString::CK_CurrentParameter: |
| 1228 | case CodeCompletionString::CK_Placeholder: |
| 1229 | processParameterChunk(ChunkText: Chunk.Text, Signature); |
| 1230 | Signal.NumberOfOptionalParameters++; |
| 1231 | break; |
| 1232 | default: |
| 1233 | Signature.label += Chunk.Text; |
| 1234 | break; |
| 1235 | } |
| 1236 | } |
| 1237 | } |
| 1238 | |
| 1239 | // FIXME(ioeric): consider moving CodeCompletionString logic here to |
| 1240 | // CompletionString.h. |
| 1241 | ScoredSignature processOverloadCandidate(const OverloadCandidate &Candidate, |
| 1242 | const CodeCompletionString &CCS, |
| 1243 | llvm::StringRef ) const { |
| 1244 | SignatureInformation Signature; |
| 1245 | SignatureQualitySignals Signal; |
| 1246 | const char *ReturnType = nullptr; |
| 1247 | |
| 1248 | markup::Document ; |
| 1249 | parseDocumentation(Input: formatDocumentation(CCS, DocComment), Output&: OverloadComment); |
| 1250 | Signature.documentation = renderDoc(Doc: OverloadComment, Kind: DocumentationFormat); |
| 1251 | Signal.Kind = Candidate.getKind(); |
| 1252 | |
| 1253 | for (const auto &Chunk : CCS) { |
| 1254 | switch (Chunk.Kind) { |
| 1255 | case CodeCompletionString::CK_ResultType: |
| 1256 | // A piece of text that describes the type of an entity or, |
| 1257 | // for functions and methods, the return type. |
| 1258 | assert(!ReturnType && "Unexpected CK_ResultType" ); |
| 1259 | ReturnType = Chunk.Text; |
| 1260 | break; |
| 1261 | case CodeCompletionString::CK_CurrentParameter: |
| 1262 | case CodeCompletionString::CK_Placeholder: |
| 1263 | processParameterChunk(ChunkText: Chunk.Text, Signature); |
| 1264 | Signal.NumberOfParameters++; |
| 1265 | break; |
| 1266 | case CodeCompletionString::CK_Optional: { |
| 1267 | // The rest of the parameters are defaulted/optional. |
| 1268 | assert(Chunk.Optional && |
| 1269 | "Expected the optional code completion string to be non-null." ); |
| 1270 | processOptionalChunk(CCS: *Chunk.Optional, Signature, Signal); |
| 1271 | break; |
| 1272 | } |
| 1273 | case CodeCompletionString::CK_VerticalSpace: |
| 1274 | break; |
| 1275 | default: |
| 1276 | Signature.label += Chunk.Text; |
| 1277 | break; |
| 1278 | } |
| 1279 | } |
| 1280 | if (ReturnType) { |
| 1281 | Signature.label += " -> " ; |
| 1282 | Signature.label += ReturnType; |
| 1283 | } |
| 1284 | dlog("Signal for {0}: {1}" , Signature, Signal); |
| 1285 | ScoredSignature Result; |
| 1286 | Result.Signature = std::move(Signature); |
| 1287 | Result.Quality = Signal; |
| 1288 | const FunctionDecl *Func = Candidate.getFunction(); |
| 1289 | if (Func && Result.Signature.documentation.value.empty()) { |
| 1290 | // Computing USR caches linkage, which may change after code completion. |
| 1291 | if (!hasUnstableLinkage(Func)) |
| 1292 | Result.IDForDoc = clangd::getSymbolID(Func); |
| 1293 | } |
| 1294 | return Result; |
| 1295 | } |
| 1296 | |
| 1297 | SignatureHelp &SigHelp; |
| 1298 | std::shared_ptr<clang::GlobalCodeCompletionAllocator> Allocator; |
| 1299 | CodeCompletionTUInfo CCTUInfo; |
| 1300 | const SymbolIndex *Index; |
| 1301 | MarkupKind DocumentationFormat; |
| 1302 | }; // SignatureHelpCollector |
| 1303 | |
| 1304 | // Used only for completion of C-style comments in function call (i.e. |
| 1305 | // /*foo=*/7). Similar to SignatureHelpCollector, but needs to do less work. |
| 1306 | class ParamNameCollector final : public CodeCompleteConsumer { |
| 1307 | public: |
| 1308 | ParamNameCollector(const clang::CodeCompleteOptions &CodeCompleteOpts, |
| 1309 | std::set<std::string> &ParamNames) |
| 1310 | : CodeCompleteConsumer(CodeCompleteOpts), |
| 1311 | Allocator(std::make_shared<clang::GlobalCodeCompletionAllocator>()), |
| 1312 | CCTUInfo(Allocator), ParamNames(ParamNames) {} |
| 1313 | |
| 1314 | void ProcessOverloadCandidates(Sema &S, unsigned CurrentArg, |
| 1315 | OverloadCandidate *Candidates, |
| 1316 | unsigned NumCandidates, |
| 1317 | SourceLocation OpenParLoc, |
| 1318 | bool Braced) override { |
| 1319 | assert(CurrentArg <= (unsigned)std::numeric_limits<int>::max() && |
| 1320 | "too many arguments" ); |
| 1321 | |
| 1322 | for (unsigned I = 0; I < NumCandidates; ++I) { |
| 1323 | if (const NamedDecl *ND = Candidates[I].getParamDecl(N: CurrentArg)) |
| 1324 | if (const auto *II = ND->getIdentifier()) |
| 1325 | ParamNames.emplace(args: II->getName()); |
| 1326 | } |
| 1327 | } |
| 1328 | |
| 1329 | private: |
| 1330 | GlobalCodeCompletionAllocator &getAllocator() override { return *Allocator; } |
| 1331 | |
| 1332 | CodeCompletionTUInfo &getCodeCompletionTUInfo() override { return CCTUInfo; } |
| 1333 | |
| 1334 | std::shared_ptr<clang::GlobalCodeCompletionAllocator> Allocator; |
| 1335 | CodeCompletionTUInfo CCTUInfo; |
| 1336 | std::set<std::string> &ParamNames; |
| 1337 | }; |
| 1338 | |
| 1339 | struct SemaCompleteInput { |
| 1340 | PathRef FileName; |
| 1341 | size_t Offset; |
| 1342 | const PreambleData &Preamble; |
| 1343 | const std::optional<PreamblePatch> Patch; |
| 1344 | const ParseInputs &ParseInput; |
| 1345 | }; |
| 1346 | |
| 1347 | void loadMainFilePreambleMacros(const Preprocessor &PP, |
| 1348 | const PreambleData &Preamble) { |
| 1349 | // The ExternalPreprocessorSource has our macros, if we know where to look. |
| 1350 | // We can read all the macros using PreambleMacros->ReadDefinedMacros(), |
| 1351 | // but this includes transitively included files, so may deserialize a lot. |
| 1352 | ExternalPreprocessorSource *PreambleMacros = PP.getExternalSource(); |
| 1353 | // As we have the names of the macros, we can look up their IdentifierInfo |
| 1354 | // and then use this to load just the macros we want. |
| 1355 | const auto &ITable = PP.getIdentifierTable(); |
| 1356 | IdentifierInfoLookup *PreambleIdentifiers = |
| 1357 | ITable.getExternalIdentifierLookup(); |
| 1358 | |
| 1359 | if (!PreambleIdentifiers || !PreambleMacros) |
| 1360 | return; |
| 1361 | for (const auto &MacroName : Preamble.Macros.Names) { |
| 1362 | if (ITable.find(MacroName.getKey()) != ITable.end()) |
| 1363 | continue; |
| 1364 | if (auto *II = PreambleIdentifiers->get(Name: MacroName.getKey())) |
| 1365 | if (II->isOutOfDate()) |
| 1366 | PreambleMacros->updateOutOfDateIdentifier(II: *II); |
| 1367 | } |
| 1368 | } |
| 1369 | |
| 1370 | // Invokes Sema code completion on a file. |
| 1371 | // If \p Includes is set, it will be updated based on the compiler invocation. |
| 1372 | bool semaCodeComplete(std::unique_ptr<CodeCompleteConsumer> Consumer, |
| 1373 | const clang::CodeCompleteOptions &Options, |
| 1374 | const SemaCompleteInput &Input, |
| 1375 | IncludeStructure *Includes = nullptr) { |
| 1376 | trace::Span Tracer("Sema completion" ); |
| 1377 | |
| 1378 | IgnoreDiagnostics IgnoreDiags; |
| 1379 | auto CI = buildCompilerInvocation(Inputs: Input.ParseInput, D&: IgnoreDiags); |
| 1380 | if (!CI) { |
| 1381 | elog(Fmt: "Couldn't create CompilerInvocation" ); |
| 1382 | return false; |
| 1383 | } |
| 1384 | auto &FrontendOpts = CI->getFrontendOpts(); |
| 1385 | FrontendOpts.SkipFunctionBodies = true; |
| 1386 | // Disable typo correction in Sema. |
| 1387 | CI->getLangOpts().SpellChecking = false; |
| 1388 | // Code completion won't trigger in delayed template bodies. |
| 1389 | // This is on-by-default in windows to allow parsing SDK headers; we're only |
| 1390 | // disabling it for the main-file (not preamble). |
| 1391 | CI->getLangOpts().DelayedTemplateParsing = false; |
| 1392 | // Setup code completion. |
| 1393 | FrontendOpts.CodeCompleteOpts = Options; |
| 1394 | FrontendOpts.CodeCompletionAt.FileName = std::string(Input.FileName); |
| 1395 | std::tie(args&: FrontendOpts.CodeCompletionAt.Line, |
| 1396 | args&: FrontendOpts.CodeCompletionAt.Column) = |
| 1397 | offsetToClangLineColumn(Code: Input.ParseInput.Contents, Offset: Input.Offset); |
| 1398 | |
| 1399 | std::unique_ptr<llvm::MemoryBuffer> ContentsBuffer = |
| 1400 | llvm::MemoryBuffer::getMemBuffer(InputData: Input.ParseInput.Contents, |
| 1401 | BufferName: Input.FileName); |
| 1402 | // The diagnostic options must be set before creating a CompilerInstance. |
| 1403 | CI->getDiagnosticOpts().IgnoreWarnings = true; |
| 1404 | // We reuse the preamble whether it's valid or not. This is a |
| 1405 | // correctness/performance tradeoff: building without a preamble is slow, and |
| 1406 | // completion is latency-sensitive. |
| 1407 | // However, if we're completing *inside* the preamble section of the draft, |
| 1408 | // overriding the preamble will break sema completion. Fortunately we can just |
| 1409 | // skip all includes in this case; these completions are really simple. |
| 1410 | PreambleBounds PreambleRegion = |
| 1411 | ComputePreambleBounds(LangOpts: CI->getLangOpts(), Buffer: *ContentsBuffer, MaxLines: 0); |
| 1412 | bool CompletingInPreamble = Input.Offset < PreambleRegion.Size || |
| 1413 | (!PreambleRegion.PreambleEndsAtStartOfLine && |
| 1414 | Input.Offset == PreambleRegion.Size); |
| 1415 | if (Input.Patch) |
| 1416 | Input.Patch->apply(CI&: *CI); |
| 1417 | // NOTE: we must call BeginSourceFile after prepareCompilerInstance. Otherwise |
| 1418 | // the remapped buffers do not get freed. |
| 1419 | llvm::IntrusiveRefCntPtr<llvm::vfs::FileSystem> VFS = |
| 1420 | Input.ParseInput.TFS->view(CWD: Input.ParseInput.CompileCommand.Directory); |
| 1421 | if (Input.Preamble.StatCache) |
| 1422 | VFS = Input.Preamble.StatCache->getConsumingFS(FS: std::move(VFS)); |
| 1423 | auto Clang = prepareCompilerInstance( |
| 1424 | std::move(CI), !CompletingInPreamble ? &Input.Preamble.Preamble : nullptr, |
| 1425 | MainFile: std::move(ContentsBuffer), std::move(VFS), IgnoreDiags); |
| 1426 | Clang->getPreprocessorOpts().SingleFileParseMode = CompletingInPreamble; |
| 1427 | Clang->setCodeCompletionConsumer(Consumer.release()); |
| 1428 | |
| 1429 | if (Input.Preamble.RequiredModules) |
| 1430 | Input.Preamble.RequiredModules->adjustHeaderSearchOptions(Options&: Clang->getHeaderSearchOpts()); |
| 1431 | |
| 1432 | SyntaxOnlyAction Action; |
| 1433 | if (!Action.BeginSourceFile(CI&: *Clang, Input: Clang->getFrontendOpts().Inputs[0])) { |
| 1434 | log(Fmt: "BeginSourceFile() failed when running codeComplete for {0}" , |
| 1435 | Vals: Input.FileName); |
| 1436 | return false; |
| 1437 | } |
| 1438 | // Macros can be defined within the preamble region of the main file. |
| 1439 | // They don't fall nicely into our index/Sema dichotomy: |
| 1440 | // - they're not indexed for completion (they're not available across files) |
| 1441 | // - but Sema code complete won't see them: as part of the preamble, they're |
| 1442 | // deserialized only when mentioned. |
| 1443 | // Force them to be deserialized so SemaCodeComplete sees them. |
| 1444 | loadMainFilePreambleMacros(PP: Clang->getPreprocessor(), Preamble: Input.Preamble); |
| 1445 | if (Includes) |
| 1446 | Includes->collect(CI: *Clang); |
| 1447 | if (llvm::Error Err = Action.Execute()) { |
| 1448 | log(Fmt: "Execute() failed when running codeComplete for {0}: {1}" , |
| 1449 | Vals: Input.FileName, Vals: toString(E: std::move(Err))); |
| 1450 | return false; |
| 1451 | } |
| 1452 | Action.EndSourceFile(); |
| 1453 | |
| 1454 | return true; |
| 1455 | } |
| 1456 | |
| 1457 | // Should we allow index completions in the specified context? |
| 1458 | bool allowIndex(CodeCompletionContext &CC) { |
| 1459 | if (!contextAllowsIndex(K: CC.getKind())) |
| 1460 | return false; |
| 1461 | // We also avoid ClassName::bar (but allow namespace::bar). |
| 1462 | auto Scope = CC.getCXXScopeSpecifier(); |
| 1463 | if (!Scope) |
| 1464 | return true; |
| 1465 | NestedNameSpecifier *NameSpec = (*Scope)->getScopeRep(); |
| 1466 | if (!NameSpec) |
| 1467 | return true; |
| 1468 | // We only query the index when qualifier is a namespace. |
| 1469 | // If it's a class, we rely solely on sema completions. |
| 1470 | switch (NameSpec->getKind()) { |
| 1471 | case NestedNameSpecifier::Global: |
| 1472 | case NestedNameSpecifier::Namespace: |
| 1473 | case NestedNameSpecifier::NamespaceAlias: |
| 1474 | return true; |
| 1475 | case NestedNameSpecifier::Super: |
| 1476 | case NestedNameSpecifier::TypeSpec: |
| 1477 | // Unresolved inside a template. |
| 1478 | case NestedNameSpecifier::Identifier: |
| 1479 | return false; |
| 1480 | } |
| 1481 | llvm_unreachable("invalid NestedNameSpecifier kind" ); |
| 1482 | } |
| 1483 | |
| 1484 | // Should we include a symbol from the index given the completion kind? |
| 1485 | // FIXME: Ideally we can filter in the fuzzy find request itself. |
| 1486 | bool includeSymbolFromIndex(CodeCompletionContext::Kind Kind, |
| 1487 | const Symbol &Sym) { |
| 1488 | // Objective-C protocols are only useful in ObjC protocol completions, |
| 1489 | // in other places they're confusing, especially when they share the same |
| 1490 | // identifier with a class. |
| 1491 | if (Sym.SymInfo.Kind == index::SymbolKind::Protocol && |
| 1492 | Sym.SymInfo.Lang == index::SymbolLanguage::ObjC) |
| 1493 | return Kind == CodeCompletionContext::CCC_ObjCProtocolName; |
| 1494 | else if (Kind == CodeCompletionContext::CCC_ObjCProtocolName) |
| 1495 | // Don't show anything else in ObjC protocol completions. |
| 1496 | return false; |
| 1497 | |
| 1498 | if (Kind == CodeCompletionContext::CCC_ObjCClassForwardDecl) |
| 1499 | return Sym.SymInfo.Kind == index::SymbolKind::Class && |
| 1500 | Sym.SymInfo.Lang == index::SymbolLanguage::ObjC; |
| 1501 | return true; |
| 1502 | } |
| 1503 | |
| 1504 | std::future<std::pair<bool, SymbolSlab>> |
| 1505 | startAsyncFuzzyFind(const SymbolIndex &Index, const FuzzyFindRequest &Req) { |
| 1506 | return runAsync<std::pair<bool, SymbolSlab>>(Action: [&Index, Req]() { |
| 1507 | trace::Span Tracer("Async fuzzyFind" ); |
| 1508 | SymbolSlab::Builder Syms; |
| 1509 | bool Incomplete = |
| 1510 | Index.fuzzyFind(Req, Callback: [&Syms](const Symbol &Sym) { Syms.insert(S: Sym); }); |
| 1511 | return std::make_pair(x&: Incomplete, y: std::move(Syms).build()); |
| 1512 | }); |
| 1513 | } |
| 1514 | |
| 1515 | // Creates a `FuzzyFindRequest` based on the cached index request from the |
| 1516 | // last completion, if any, and the speculated completion filter text in the |
| 1517 | // source code. |
| 1518 | FuzzyFindRequest speculativeFuzzyFindRequestForCompletion( |
| 1519 | FuzzyFindRequest CachedReq, const CompletionPrefix &HeuristicPrefix) { |
| 1520 | CachedReq.Query = std::string(HeuristicPrefix.Name); |
| 1521 | return CachedReq; |
| 1522 | } |
| 1523 | |
| 1524 | // This function is similar to Lexer::findNextToken(), but assumes |
| 1525 | // that the input SourceLocation is the completion point (which is |
| 1526 | // a case findNextToken() does not handle). |
| 1527 | std::optional<Token> |
| 1528 | findTokenAfterCompletionPoint(SourceLocation CompletionPoint, |
| 1529 | const SourceManager &SM, |
| 1530 | const LangOptions &LangOpts) { |
| 1531 | SourceLocation Loc = CompletionPoint; |
| 1532 | if (Loc.isMacroID()) { |
| 1533 | if (!Lexer::isAtEndOfMacroExpansion(loc: Loc, SM, LangOpts, MacroEnd: &Loc)) |
| 1534 | return std::nullopt; |
| 1535 | } |
| 1536 | |
| 1537 | // Advance to the next SourceLocation after the completion point. |
| 1538 | // Lexer::findNextToken() would call MeasureTokenLength() here, |
| 1539 | // which does not handle the completion point (and can't, because |
| 1540 | // the Lexer instance it constructs internally doesn't have a |
| 1541 | // Preprocessor and so doesn't know about the completion point). |
| 1542 | Loc = Loc.getLocWithOffset(Offset: 1); |
| 1543 | |
| 1544 | // Break down the source location. |
| 1545 | std::pair<FileID, unsigned> LocInfo = SM.getDecomposedLoc(Loc); |
| 1546 | |
| 1547 | // Try to load the file buffer. |
| 1548 | bool InvalidTemp = false; |
| 1549 | StringRef File = SM.getBufferData(FID: LocInfo.first, Invalid: &InvalidTemp); |
| 1550 | if (InvalidTemp) |
| 1551 | return std::nullopt; |
| 1552 | |
| 1553 | const char *TokenBegin = File.data() + LocInfo.second; |
| 1554 | |
| 1555 | // Lex from the start of the given location. |
| 1556 | Lexer TheLexer(SM.getLocForStartOfFile(FID: LocInfo.first), LangOpts, File.begin(), |
| 1557 | TokenBegin, File.end()); |
| 1558 | // Find the token. |
| 1559 | Token Tok; |
| 1560 | TheLexer.LexFromRawLexer(Result&: Tok); |
| 1561 | return Tok; |
| 1562 | } |
| 1563 | |
| 1564 | // Runs Sema-based (AST) and Index-based completion, returns merged results. |
| 1565 | // |
| 1566 | // There are a few tricky considerations: |
| 1567 | // - the AST provides information needed for the index query (e.g. which |
| 1568 | // namespaces to search in). So Sema must start first. |
| 1569 | // - we only want to return the top results (Opts.Limit). |
| 1570 | // Building CompletionItems for everything else is wasteful, so we want to |
| 1571 | // preserve the "native" format until we're done with scoring. |
| 1572 | // - the data underlying Sema completion items is owned by the AST and various |
| 1573 | // other arenas, which must stay alive for us to build CompletionItems. |
| 1574 | // - we may get duplicate results from Sema and the Index, we need to merge. |
| 1575 | // |
| 1576 | // So we start Sema completion first, and do all our work in its callback. |
| 1577 | // We use the Sema context information to query the index. |
| 1578 | // Then we merge the two result sets, producing items that are Sema/Index/Both. |
| 1579 | // These items are scored, and the top N are synthesized into the LSP response. |
| 1580 | // Finally, we can clean up the data structures created by Sema completion. |
| 1581 | // |
| 1582 | // Main collaborators are: |
| 1583 | // - semaCodeComplete sets up the compiler machinery to run code completion. |
| 1584 | // - CompletionRecorder captures Sema completion results, including context. |
| 1585 | // - SymbolIndex (Opts.Index) provides index completion results as Symbols |
| 1586 | // - CompletionCandidates are the result of merging Sema and Index results. |
| 1587 | // Each candidate points to an underlying CodeCompletionResult (Sema), a |
| 1588 | // Symbol (Index), or both. It computes the result quality score. |
| 1589 | // CompletionCandidate also does conversion to CompletionItem (at the end). |
| 1590 | // - FuzzyMatcher scores how the candidate matches the partial identifier. |
| 1591 | // This score is combined with the result quality score for the final score. |
| 1592 | // - TopN determines the results with the best score. |
| 1593 | class CodeCompleteFlow { |
| 1594 | PathRef FileName; |
| 1595 | IncludeStructure Includes; // Complete once the compiler runs. |
| 1596 | SpeculativeFuzzyFind *SpecFuzzyFind; // Can be nullptr. |
| 1597 | const CodeCompleteOptions &Opts; |
| 1598 | |
| 1599 | // Sema takes ownership of Recorder. Recorder is valid until Sema cleanup. |
| 1600 | CompletionRecorder *Recorder = nullptr; |
| 1601 | CodeCompletionContext::Kind CCContextKind = CodeCompletionContext::CCC_Other; |
| 1602 | bool IsUsingDeclaration = false; |
| 1603 | // The snippets will not be generated if the token following completion |
| 1604 | // location is an opening parenthesis (tok::l_paren) because this would add |
| 1605 | // extra parenthesis. |
| 1606 | tok::TokenKind NextTokenKind = tok::eof; |
| 1607 | // Counters for logging. |
| 1608 | int NSema = 0, NIndex = 0, NSemaAndIndex = 0, NIdent = 0; |
| 1609 | bool Incomplete = false; // Would more be available with a higher limit? |
| 1610 | CompletionPrefix HeuristicPrefix; |
| 1611 | std::optional<FuzzyMatcher> Filter; // Initialized once Sema runs. |
| 1612 | Range ReplacedRange; |
| 1613 | std::vector<std::string> QueryScopes; // Initialized once Sema runs. |
| 1614 | std::vector<std::string> AccessibleScopes; // Initialized once Sema runs. |
| 1615 | // Initialized once QueryScopes is initialized, if there are scopes. |
| 1616 | std::optional<ScopeDistance> ScopeProximity; |
| 1617 | std::optional<OpaqueType> PreferredType; // Initialized once Sema runs. |
| 1618 | // Whether to query symbols from any scope. Initialized once Sema runs. |
| 1619 | bool AllScopes = false; |
| 1620 | llvm::StringSet<> ContextWords; |
| 1621 | // Include-insertion and proximity scoring rely on the include structure. |
| 1622 | // This is available after Sema has run. |
| 1623 | std::optional<IncludeInserter> Inserter; // Available during runWithSema. |
| 1624 | std::optional<URIDistance> FileProximity; // Initialized once Sema runs. |
| 1625 | /// Speculative request based on the cached request and the filter text before |
| 1626 | /// the cursor. |
| 1627 | /// Initialized right before sema run. This is only set if `SpecFuzzyFind` is |
| 1628 | /// set and contains a cached request. |
| 1629 | std::optional<FuzzyFindRequest> SpecReq; |
| 1630 | |
| 1631 | public: |
| 1632 | // A CodeCompleteFlow object is only useful for calling run() exactly once. |
| 1633 | CodeCompleteFlow(PathRef FileName, const IncludeStructure &Includes, |
| 1634 | SpeculativeFuzzyFind *SpecFuzzyFind, |
| 1635 | const CodeCompleteOptions &Opts) |
| 1636 | : FileName(FileName), Includes(Includes), SpecFuzzyFind(SpecFuzzyFind), |
| 1637 | Opts(Opts) {} |
| 1638 | |
| 1639 | CodeCompleteResult run(const SemaCompleteInput &SemaCCInput) && { |
| 1640 | trace::Span Tracer("CodeCompleteFlow" ); |
| 1641 | HeuristicPrefix = guessCompletionPrefix(Content: SemaCCInput.ParseInput.Contents, |
| 1642 | Offset: SemaCCInput.Offset); |
| 1643 | populateContextWords(Content: SemaCCInput.ParseInput.Contents); |
| 1644 | if (Opts.Index && SpecFuzzyFind && SpecFuzzyFind->CachedReq) { |
| 1645 | assert(!SpecFuzzyFind->Result.valid()); |
| 1646 | SpecReq = speculativeFuzzyFindRequestForCompletion( |
| 1647 | CachedReq: *SpecFuzzyFind->CachedReq, HeuristicPrefix); |
| 1648 | SpecFuzzyFind->Result = startAsyncFuzzyFind(Index: *Opts.Index, Req: *SpecReq); |
| 1649 | } |
| 1650 | |
| 1651 | // We run Sema code completion first. It builds an AST and calculates: |
| 1652 | // - completion results based on the AST. |
| 1653 | // - partial identifier and context. We need these for the index query. |
| 1654 | CodeCompleteResult Output; |
| 1655 | auto RecorderOwner = std::make_unique<CompletionRecorder>(args: Opts, args: [&]() { |
| 1656 | assert(Recorder && "Recorder is not set" ); |
| 1657 | CCContextKind = Recorder->CCContext.getKind(); |
| 1658 | IsUsingDeclaration = Recorder->CCContext.isUsingDeclaration(); |
| 1659 | auto Style = getFormatStyleForFile(File: SemaCCInput.FileName, |
| 1660 | Content: SemaCCInput.ParseInput.Contents, |
| 1661 | TFS: *SemaCCInput.ParseInput.TFS, FormatFile: false); |
| 1662 | const auto NextToken = findTokenAfterCompletionPoint( |
| 1663 | CompletionPoint: Recorder->CCSema->getPreprocessor().getCodeCompletionLoc(), |
| 1664 | SM: Recorder->CCSema->getSourceManager(), LangOpts: Recorder->CCSema->LangOpts); |
| 1665 | if (NextToken) |
| 1666 | NextTokenKind = NextToken->getKind(); |
| 1667 | // If preprocessor was run, inclusions from preprocessor callback should |
| 1668 | // already be added to Includes. |
| 1669 | Inserter.emplace( |
| 1670 | args: SemaCCInput.FileName, args: SemaCCInput.ParseInput.Contents, args&: Style, |
| 1671 | args: SemaCCInput.ParseInput.CompileCommand.Directory, |
| 1672 | args: &Recorder->CCSema->getPreprocessor().getHeaderSearchInfo(), |
| 1673 | args: Config::current().Style.QuotedHeaders, |
| 1674 | args: Config::current().Style.AngledHeaders); |
| 1675 | for (const auto &Inc : Includes.MainFileIncludes) |
| 1676 | Inserter->addExisting(Inc); |
| 1677 | |
| 1678 | // Most of the cost of file proximity is in initializing the FileDistance |
| 1679 | // structures based on the observed includes, once per query. Conceptually |
| 1680 | // that happens here (though the per-URI-scheme initialization is lazy). |
| 1681 | // The per-result proximity scoring is (amortized) very cheap. |
| 1682 | FileDistanceOptions ProxOpts{}; // Use defaults. |
| 1683 | const auto &SM = Recorder->CCSema->getSourceManager(); |
| 1684 | llvm::StringMap<SourceParams> ProxSources; |
| 1685 | auto MainFileID = |
| 1686 | Includes.getID(Entry: SM.getFileEntryForID(FID: SM.getMainFileID())); |
| 1687 | assert(MainFileID); |
| 1688 | for (auto &HeaderIDAndDepth : Includes.includeDepth(Root: *MainFileID)) { |
| 1689 | auto &Source = |
| 1690 | ProxSources[Includes.getRealPath(ID: HeaderIDAndDepth.getFirst())]; |
| 1691 | Source.Cost = HeaderIDAndDepth.getSecond() * ProxOpts.IncludeCost; |
| 1692 | // Symbols near our transitive includes are good, but only consider |
| 1693 | // things in the same directory or below it. Otherwise there can be |
| 1694 | // many false positives. |
| 1695 | if (HeaderIDAndDepth.getSecond() > 0) |
| 1696 | Source.MaxUpTraversals = 1; |
| 1697 | } |
| 1698 | FileProximity.emplace(args&: ProxSources, args&: ProxOpts); |
| 1699 | |
| 1700 | Output = runWithSema(); |
| 1701 | Inserter.reset(); // Make sure this doesn't out-live Clang. |
| 1702 | SPAN_ATTACH(Tracer, "sema_completion_kind" , |
| 1703 | getCompletionKindString(CCContextKind)); |
| 1704 | log("Code complete: sema context {0}, query scopes [{1}] (AnyScope={2}), " |
| 1705 | "expected type {3}{4}" , |
| 1706 | getCompletionKindString(Kind: CCContextKind), |
| 1707 | llvm::join(Begin: QueryScopes.begin(), End: QueryScopes.end(), Separator: "," ), AllScopes, |
| 1708 | PreferredType ? Recorder->CCContext.getPreferredType().getAsString() |
| 1709 | : "<none>" , |
| 1710 | IsUsingDeclaration ? ", inside using declaration" : "" ); |
| 1711 | }); |
| 1712 | |
| 1713 | Recorder = RecorderOwner.get(); |
| 1714 | |
| 1715 | semaCodeComplete(std::move(RecorderOwner), Opts.getClangCompleteOpts(), |
| 1716 | SemaCCInput, &Includes); |
| 1717 | logResults(Output, Tracer); |
| 1718 | return Output; |
| 1719 | } |
| 1720 | |
| 1721 | void logResults(const CodeCompleteResult &Output, const trace::Span &Tracer) { |
| 1722 | SPAN_ATTACH(Tracer, "sema_results" , NSema); |
| 1723 | SPAN_ATTACH(Tracer, "index_results" , NIndex); |
| 1724 | SPAN_ATTACH(Tracer, "merged_results" , NSemaAndIndex); |
| 1725 | SPAN_ATTACH(Tracer, "identifier_results" , NIdent); |
| 1726 | SPAN_ATTACH(Tracer, "returned_results" , int64_t(Output.Completions.size())); |
| 1727 | SPAN_ATTACH(Tracer, "incomplete" , Output.HasMore); |
| 1728 | log(Fmt: "Code complete: {0} results from Sema, {1} from Index, " |
| 1729 | "{2} matched, {3} from identifiers, {4} returned{5}." , |
| 1730 | Vals&: NSema, Vals&: NIndex, Vals&: NSemaAndIndex, Vals&: NIdent, Vals: Output.Completions.size(), |
| 1731 | Vals: Output.HasMore ? " (incomplete)" : "" ); |
| 1732 | assert(!Opts.Limit || Output.Completions.size() <= Opts.Limit); |
| 1733 | // We don't assert that isIncomplete means we hit a limit. |
| 1734 | // Indexes may choose to impose their own limits even if we don't have one. |
| 1735 | } |
| 1736 | |
| 1737 | CodeCompleteResult runWithoutSema(llvm::StringRef Content, size_t Offset, |
| 1738 | const ThreadsafeFS &TFS) && { |
| 1739 | trace::Span Tracer("CodeCompleteWithoutSema" ); |
| 1740 | // Fill in fields normally set by runWithSema() |
| 1741 | HeuristicPrefix = guessCompletionPrefix(Content, Offset); |
| 1742 | populateContextWords(Content); |
| 1743 | CCContextKind = CodeCompletionContext::CCC_Recovery; |
| 1744 | IsUsingDeclaration = false; |
| 1745 | Filter = FuzzyMatcher(HeuristicPrefix.Name); |
| 1746 | auto Pos = offsetToPosition(Code: Content, Offset); |
| 1747 | ReplacedRange.start = ReplacedRange.end = Pos; |
| 1748 | ReplacedRange.start.character -= HeuristicPrefix.Name.size(); |
| 1749 | |
| 1750 | llvm::StringMap<SourceParams> ProxSources; |
| 1751 | ProxSources[FileName].Cost = 0; |
| 1752 | FileProximity.emplace(args&: ProxSources); |
| 1753 | |
| 1754 | auto Style = getFormatStyleForFile(File: FileName, Content, TFS, FormatFile: false); |
| 1755 | // This will only insert verbatim headers. |
| 1756 | Inserter.emplace(args&: FileName, args&: Content, args&: Style, |
| 1757 | /*BuildDir=*/args: "" , /*HeaderSearchInfo=*/args: nullptr, |
| 1758 | args: Config::current().Style.QuotedHeaders, |
| 1759 | args: Config::current().Style.AngledHeaders); |
| 1760 | |
| 1761 | auto Identifiers = collectIdentifiers(Content, Style); |
| 1762 | std::vector<RawIdentifier> IdentifierResults; |
| 1763 | for (const auto &IDAndCount : Identifiers) { |
| 1764 | RawIdentifier ID; |
| 1765 | ID.Name = IDAndCount.first(); |
| 1766 | ID.References = IDAndCount.second; |
| 1767 | // Avoid treating typed filter as an identifier. |
| 1768 | if (ID.Name == HeuristicPrefix.Name) |
| 1769 | --ID.References; |
| 1770 | if (ID.References > 0) |
| 1771 | IdentifierResults.push_back(x: std::move(ID)); |
| 1772 | } |
| 1773 | |
| 1774 | // Simplified version of getQueryScopes(): |
| 1775 | // - accessible scopes are determined heuristically. |
| 1776 | // - all-scopes query if no qualifier was typed (and it's allowed). |
| 1777 | SpecifiedScope Scopes; |
| 1778 | Scopes.QueryScopes = visibleNamespaces( |
| 1779 | Code: Content.take_front(N: Offset), LangOpts: format::getFormattingLangOpts(Style)); |
| 1780 | for (std::string &S : Scopes.QueryScopes) |
| 1781 | if (!S.empty()) |
| 1782 | S.append(s: "::" ); // visibleNamespaces doesn't include trailing ::. |
| 1783 | if (HeuristicPrefix.Qualifier.empty()) |
| 1784 | AllScopes = Opts.AllScopes; |
| 1785 | else if (HeuristicPrefix.Qualifier.starts_with(Prefix: "::" )) { |
| 1786 | Scopes.QueryScopes = {"" }; |
| 1787 | Scopes.UnresolvedQualifier = |
| 1788 | std::string(HeuristicPrefix.Qualifier.drop_front(N: 2)); |
| 1789 | } else |
| 1790 | Scopes.UnresolvedQualifier = std::string(HeuristicPrefix.Qualifier); |
| 1791 | // First scope is the (modified) enclosing scope. |
| 1792 | QueryScopes = Scopes.scopesForIndexQuery(); |
| 1793 | AccessibleScopes = QueryScopes; |
| 1794 | ScopeProximity.emplace(args&: QueryScopes); |
| 1795 | |
| 1796 | SymbolSlab IndexResults = Opts.Index ? queryIndex() : SymbolSlab(); |
| 1797 | |
| 1798 | CodeCompleteResult Output = toCodeCompleteResult(Scored: mergeResults( |
| 1799 | /*SemaResults=*/{}, IndexResults, IdentifierResults)); |
| 1800 | Output.RanParser = false; |
| 1801 | logResults(Output, Tracer); |
| 1802 | return Output; |
| 1803 | } |
| 1804 | |
| 1805 | private: |
| 1806 | void populateContextWords(llvm::StringRef Content) { |
| 1807 | // Take last 3 lines before the completion point. |
| 1808 | unsigned RangeEnd = HeuristicPrefix.Qualifier.begin() - Content.data(), |
| 1809 | RangeBegin = RangeEnd; |
| 1810 | for (size_t I = 0; I < 3 && RangeBegin > 0; ++I) { |
| 1811 | auto PrevNL = Content.rfind(C: '\n', From: RangeBegin); |
| 1812 | if (PrevNL == StringRef::npos) { |
| 1813 | RangeBegin = 0; |
| 1814 | break; |
| 1815 | } |
| 1816 | RangeBegin = PrevNL; |
| 1817 | } |
| 1818 | |
| 1819 | ContextWords = collectWords(Content: Content.slice(Start: RangeBegin, End: RangeEnd)); |
| 1820 | dlog("Completion context words: {0}" , |
| 1821 | llvm::join(ContextWords.keys(), ", " )); |
| 1822 | } |
| 1823 | |
| 1824 | // This is called by run() once Sema code completion is done, but before the |
| 1825 | // Sema data structures are torn down. It does all the real work. |
| 1826 | CodeCompleteResult runWithSema() { |
| 1827 | const auto &CodeCompletionRange = CharSourceRange::getCharRange( |
| 1828 | R: Recorder->CCSema->getPreprocessor().getCodeCompletionTokenRange()); |
| 1829 | // When we are getting completions with an empty identifier, for example |
| 1830 | // std::vector<int> asdf; |
| 1831 | // asdf.^; |
| 1832 | // Then the range will be invalid and we will be doing insertion, use |
| 1833 | // current cursor position in such cases as range. |
| 1834 | if (CodeCompletionRange.isValid()) { |
| 1835 | ReplacedRange = halfOpenToRange(SM: Recorder->CCSema->getSourceManager(), |
| 1836 | R: CodeCompletionRange); |
| 1837 | } else { |
| 1838 | const auto &Pos = sourceLocToPosition( |
| 1839 | SM: Recorder->CCSema->getSourceManager(), |
| 1840 | Loc: Recorder->CCSema->getPreprocessor().getCodeCompletionLoc()); |
| 1841 | ReplacedRange.start = ReplacedRange.end = Pos; |
| 1842 | } |
| 1843 | Filter = FuzzyMatcher( |
| 1844 | Recorder->CCSema->getPreprocessor().getCodeCompletionFilter()); |
| 1845 | auto SpecifiedScopes = getQueryScopes( |
| 1846 | Recorder->CCContext, *Recorder->CCSema, HeuristicPrefix, Opts); |
| 1847 | |
| 1848 | QueryScopes = SpecifiedScopes.scopesForIndexQuery(); |
| 1849 | AccessibleScopes = SpecifiedScopes.scopesForQualification(); |
| 1850 | AllScopes = SpecifiedScopes.AllowAllScopes; |
| 1851 | if (!QueryScopes.empty()) |
| 1852 | ScopeProximity.emplace(args&: QueryScopes); |
| 1853 | PreferredType = |
| 1854 | OpaqueType::fromType(Ctx&: Recorder->CCSema->getASTContext(), |
| 1855 | Type: Recorder->CCContext.getPreferredType()); |
| 1856 | // Sema provides the needed context to query the index. |
| 1857 | // FIXME: in addition to querying for extra/overlapping symbols, we should |
| 1858 | // explicitly request symbols corresponding to Sema results. |
| 1859 | // We can use their signals even if the index can't suggest them. |
| 1860 | // We must copy index results to preserve them, but there are at most Limit. |
| 1861 | auto IndexResults = (Opts.Index && allowIndex(Recorder->CCContext)) |
| 1862 | ? queryIndex() |
| 1863 | : SymbolSlab(); |
| 1864 | trace::Span Tracer("Populate CodeCompleteResult" ); |
| 1865 | // Merge Sema and Index results, score them, and pick the winners. |
| 1866 | auto Top = |
| 1867 | mergeResults(SemaResults: Recorder->Results, IndexResults: IndexResults, /*Identifiers*/ IdentifierResults: {}); |
| 1868 | return toCodeCompleteResult(Scored: Top); |
| 1869 | } |
| 1870 | |
| 1871 | CodeCompleteResult |
| 1872 | toCodeCompleteResult(const std::vector<ScoredBundle> &Scored) { |
| 1873 | CodeCompleteResult Output; |
| 1874 | |
| 1875 | // Convert the results to final form, assembling the expensive strings. |
| 1876 | // If necessary, search the index for documentation comments. |
| 1877 | LookupRequest Req; |
| 1878 | llvm::DenseMap<SymbolID, uint32_t> SymbolToCompletion; |
| 1879 | for (auto &C : Scored) { |
| 1880 | Output.Completions.push_back(x: toCodeCompletion(Bundle: C.first)); |
| 1881 | Output.Completions.back().Score = C.second; |
| 1882 | Output.Completions.back().CompletionTokenRange = ReplacedRange; |
| 1883 | if (Opts.Index && !Output.Completions.back().Documentation) { |
| 1884 | for (auto &Cand : C.first) { |
| 1885 | if (Cand.SemaResult && |
| 1886 | Cand.SemaResult->Kind == CodeCompletionResult::RK_Declaration) { |
| 1887 | auto ID = clangd::getSymbolID(Cand.SemaResult->getDeclaration()); |
| 1888 | if (!ID) |
| 1889 | continue; |
| 1890 | Req.IDs.insert(ID); |
| 1891 | SymbolToCompletion[ID] = Output.Completions.size() - 1; |
| 1892 | } |
| 1893 | } |
| 1894 | } |
| 1895 | } |
| 1896 | Output.HasMore = Incomplete; |
| 1897 | Output.Context = CCContextKind; |
| 1898 | Output.CompletionRange = ReplacedRange; |
| 1899 | |
| 1900 | // Look up documentation from the index. |
| 1901 | if (Opts.Index) { |
| 1902 | Opts.Index->lookup(Req, Callback: [&](const Symbol &S) { |
| 1903 | if (S.Documentation.empty()) |
| 1904 | return; |
| 1905 | auto &C = Output.Completions[SymbolToCompletion.at(Val: S.ID)]; |
| 1906 | C.Documentation.emplace(); |
| 1907 | parseDocumentation(Input: S.Documentation, Output&: *C.Documentation); |
| 1908 | }); |
| 1909 | } |
| 1910 | |
| 1911 | return Output; |
| 1912 | } |
| 1913 | |
| 1914 | SymbolSlab queryIndex() { |
| 1915 | trace::Span Tracer("Query index" ); |
| 1916 | SPAN_ATTACH(Tracer, "limit" , int64_t(Opts.Limit)); |
| 1917 | |
| 1918 | // Build the query. |
| 1919 | FuzzyFindRequest Req; |
| 1920 | if (Opts.Limit) |
| 1921 | Req.Limit = Opts.Limit; |
| 1922 | Req.Query = std::string(Filter->pattern()); |
| 1923 | Req.RestrictForCodeCompletion = true; |
| 1924 | Req.Scopes = QueryScopes; |
| 1925 | Req.AnyScope = AllScopes; |
| 1926 | // FIXME: we should send multiple weighted paths here. |
| 1927 | Req.ProximityPaths.push_back(x: std::string(FileName)); |
| 1928 | if (PreferredType) |
| 1929 | Req.PreferredTypes.push_back(x: std::string(PreferredType->raw())); |
| 1930 | vlog(Fmt: "Code complete: fuzzyFind({0:2})" , Vals: toJSON(Request: Req)); |
| 1931 | |
| 1932 | if (SpecFuzzyFind) |
| 1933 | SpecFuzzyFind->NewReq = Req; |
| 1934 | if (SpecFuzzyFind && SpecFuzzyFind->Result.valid() && (*SpecReq == Req)) { |
| 1935 | vlog(Fmt: "Code complete: speculative fuzzy request matches the actual index " |
| 1936 | "request. Waiting for the speculative index results." ); |
| 1937 | SPAN_ATTACH(Tracer, "Speculative results" , true); |
| 1938 | |
| 1939 | trace::Span WaitSpec("Wait speculative results" ); |
| 1940 | auto SpecRes = SpecFuzzyFind->Result.get(); |
| 1941 | Incomplete |= SpecRes.first; |
| 1942 | return std::move(SpecRes.second); |
| 1943 | } |
| 1944 | |
| 1945 | SPAN_ATTACH(Tracer, "Speculative results" , false); |
| 1946 | |
| 1947 | // Run the query against the index. |
| 1948 | SymbolSlab::Builder ResultsBuilder; |
| 1949 | Incomplete |= Opts.Index->fuzzyFind( |
| 1950 | Req, Callback: [&](const Symbol &Sym) { ResultsBuilder.insert(S: Sym); }); |
| 1951 | return std::move(ResultsBuilder).build(); |
| 1952 | } |
| 1953 | |
| 1954 | // Merges Sema and Index results where possible, to form CompletionCandidates. |
| 1955 | // \p Identifiers is raw identifiers that can also be completion candidates. |
| 1956 | // Identifiers are not merged with results from index or sema. |
| 1957 | // Groups overloads if desired, to form CompletionCandidate::Bundles. The |
| 1958 | // bundles are scored and top results are returned, best to worst. |
| 1959 | std::vector<ScoredBundle> |
| 1960 | mergeResults(const std::vector<CodeCompletionResult> &SemaResults, |
| 1961 | const SymbolSlab &IndexResults, |
| 1962 | const std::vector<RawIdentifier> &IdentifierResults) { |
| 1963 | trace::Span Tracer("Merge and score results" ); |
| 1964 | std::vector<CompletionCandidate::Bundle> Bundles; |
| 1965 | llvm::DenseMap<size_t, size_t> BundleLookup; |
| 1966 | auto AddToBundles = [&](const CodeCompletionResult *SemaResult, |
| 1967 | const Symbol *IndexResult, |
| 1968 | const RawIdentifier *IdentifierResult) { |
| 1969 | CompletionCandidate C; |
| 1970 | C.SemaResult = SemaResult; |
| 1971 | C.IndexResult = IndexResult; |
| 1972 | C.IdentifierResult = IdentifierResult; |
| 1973 | if (C.IndexResult) { |
| 1974 | C.Name = IndexResult->Name; |
| 1975 | C.RankedIncludeHeaders = getRankedIncludes(Sym: *C.IndexResult); |
| 1976 | } else if (C.SemaResult) { |
| 1977 | C.Name = Recorder->getName(Result: *SemaResult); |
| 1978 | } else { |
| 1979 | assert(IdentifierResult); |
| 1980 | C.Name = IdentifierResult->Name; |
| 1981 | } |
| 1982 | if (auto OverloadSet = C.overloadSet( |
| 1983 | Opts, FileName, Inserter: Inserter ? &*Inserter : nullptr, CCContextKind)) { |
| 1984 | auto Ret = BundleLookup.try_emplace(Key: OverloadSet, Args: Bundles.size()); |
| 1985 | if (Ret.second) |
| 1986 | Bundles.emplace_back(); |
| 1987 | Bundles[Ret.first->second].push_back(Elt: std::move(C)); |
| 1988 | } else { |
| 1989 | Bundles.emplace_back(); |
| 1990 | Bundles.back().push_back(Elt: std::move(C)); |
| 1991 | } |
| 1992 | }; |
| 1993 | llvm::DenseSet<const Symbol *> UsedIndexResults; |
| 1994 | auto CorrespondingIndexResult = |
| 1995 | [&](const CodeCompletionResult &SemaResult) -> const Symbol * { |
| 1996 | if (auto SymID = |
| 1997 | getSymbolID(R: SemaResult, SM: Recorder->CCSema->getSourceManager())) { |
| 1998 | auto I = IndexResults.find(SymID); |
| 1999 | if (I != IndexResults.end()) { |
| 2000 | UsedIndexResults.insert(V: &*I); |
| 2001 | return &*I; |
| 2002 | } |
| 2003 | } |
| 2004 | return nullptr; |
| 2005 | }; |
| 2006 | // Emit all Sema results, merging them with Index results if possible. |
| 2007 | for (auto &SemaResult : SemaResults) |
| 2008 | AddToBundles(&SemaResult, CorrespondingIndexResult(SemaResult), nullptr); |
| 2009 | // Now emit any Index-only results. |
| 2010 | for (const auto &IndexResult : IndexResults) { |
| 2011 | if (UsedIndexResults.count(V: &IndexResult)) |
| 2012 | continue; |
| 2013 | if (!includeSymbolFromIndex(Kind: CCContextKind, Sym: IndexResult)) |
| 2014 | continue; |
| 2015 | AddToBundles(/*SemaResult=*/nullptr, &IndexResult, nullptr); |
| 2016 | } |
| 2017 | // Emit identifier results. |
| 2018 | for (const auto &Ident : IdentifierResults) |
| 2019 | AddToBundles(/*SemaResult=*/nullptr, /*IndexResult=*/nullptr, &Ident); |
| 2020 | // We only keep the best N results at any time, in "native" format. |
| 2021 | TopN<ScoredBundle, ScoredBundleGreater> Top( |
| 2022 | Opts.Limit == 0 ? std::numeric_limits<size_t>::max() : Opts.Limit); |
| 2023 | for (auto &Bundle : Bundles) |
| 2024 | addCandidate(Candidates&: Top, Bundle: std::move(Bundle)); |
| 2025 | return std::move(Top).items(); |
| 2026 | } |
| 2027 | |
| 2028 | std::optional<float> fuzzyScore(const CompletionCandidate &C) { |
| 2029 | // Macros can be very spammy, so we only support prefix completion. |
| 2030 | if (((C.SemaResult && |
| 2031 | C.SemaResult->Kind == CodeCompletionResult::RK_Macro) || |
| 2032 | (C.IndexResult && |
| 2033 | C.IndexResult->SymInfo.Kind == index::SymbolKind::Macro)) && |
| 2034 | !C.Name.starts_with_insensitive(Prefix: Filter->pattern())) |
| 2035 | return std::nullopt; |
| 2036 | return Filter->match(Word: C.Name); |
| 2037 | } |
| 2038 | |
| 2039 | CodeCompletion::Scores |
| 2040 | evaluateCompletion(const SymbolQualitySignals &Quality, |
| 2041 | const SymbolRelevanceSignals &Relevance) { |
| 2042 | using RM = CodeCompleteOptions::CodeCompletionRankingModel; |
| 2043 | CodeCompletion::Scores Scores; |
| 2044 | switch (Opts.RankingModel) { |
| 2045 | case RM::Heuristics: |
| 2046 | Scores.Quality = Quality.evaluateHeuristics(); |
| 2047 | Scores.Relevance = Relevance.evaluateHeuristics(); |
| 2048 | Scores.Total = |
| 2049 | evaluateSymbolAndRelevance(SymbolQuality: Scores.Quality, SymbolRelevance: Scores.Relevance); |
| 2050 | // NameMatch is in fact a multiplier on total score, so rescoring is |
| 2051 | // sound. |
| 2052 | Scores.ExcludingName = |
| 2053 | Relevance.NameMatch > std::numeric_limits<float>::epsilon() |
| 2054 | ? Scores.Total / Relevance.NameMatch |
| 2055 | : Scores.Quality; |
| 2056 | return Scores; |
| 2057 | |
| 2058 | case RM::DecisionForest: |
| 2059 | DecisionForestScores DFScores = Opts.DecisionForestScorer( |
| 2060 | Quality, Relevance, Opts.DecisionForestBase); |
| 2061 | Scores.ExcludingName = DFScores.ExcludingName; |
| 2062 | Scores.Total = DFScores.Total; |
| 2063 | return Scores; |
| 2064 | } |
| 2065 | llvm_unreachable("Unhandled CodeCompletion ranking model." ); |
| 2066 | } |
| 2067 | |
| 2068 | // Scores a candidate and adds it to the TopN structure. |
| 2069 | void addCandidate(TopN<ScoredBundle, ScoredBundleGreater> &Candidates, |
| 2070 | CompletionCandidate::Bundle Bundle) { |
| 2071 | SymbolQualitySignals Quality; |
| 2072 | SymbolRelevanceSignals Relevance; |
| 2073 | Relevance.Context = CCContextKind; |
| 2074 | Relevance.Name = Bundle.front().Name; |
| 2075 | Relevance.FilterLength = HeuristicPrefix.Name.size(); |
| 2076 | Relevance.Query = SymbolRelevanceSignals::CodeComplete; |
| 2077 | Relevance.FileProximityMatch = &*FileProximity; |
| 2078 | if (ScopeProximity) |
| 2079 | Relevance.ScopeProximityMatch = &*ScopeProximity; |
| 2080 | if (PreferredType) |
| 2081 | Relevance.HadContextType = true; |
| 2082 | Relevance.ContextWords = &ContextWords; |
| 2083 | Relevance.MainFileSignals = Opts.MainFileSignals; |
| 2084 | |
| 2085 | auto &First = Bundle.front(); |
| 2086 | if (auto FuzzyScore = fuzzyScore(C: First)) |
| 2087 | Relevance.NameMatch = *FuzzyScore; |
| 2088 | else |
| 2089 | return; |
| 2090 | SymbolOrigin Origin = SymbolOrigin::Unknown; |
| 2091 | bool FromIndex = false; |
| 2092 | for (const auto &Candidate : Bundle) { |
| 2093 | if (Candidate.IndexResult) { |
| 2094 | Quality.merge(IndexResult: *Candidate.IndexResult); |
| 2095 | Relevance.merge(IndexResult: *Candidate.IndexResult); |
| 2096 | Origin |= Candidate.IndexResult->Origin; |
| 2097 | FromIndex = true; |
| 2098 | if (!Candidate.IndexResult->Type.empty()) |
| 2099 | Relevance.HadSymbolType |= true; |
| 2100 | if (PreferredType && |
| 2101 | PreferredType->raw() == Candidate.IndexResult->Type) { |
| 2102 | Relevance.TypeMatchesPreferred = true; |
| 2103 | } |
| 2104 | } |
| 2105 | if (Candidate.SemaResult) { |
| 2106 | Quality.merge(SemaCCResult: *Candidate.SemaResult); |
| 2107 | Relevance.merge(SemaResult: *Candidate.SemaResult); |
| 2108 | if (PreferredType) { |
| 2109 | if (auto CompletionType = OpaqueType::fromCompletionResult( |
| 2110 | Ctx&: Recorder->CCSema->getASTContext(), R: *Candidate.SemaResult)) { |
| 2111 | Relevance.HadSymbolType |= true; |
| 2112 | if (PreferredType == CompletionType) |
| 2113 | Relevance.TypeMatchesPreferred = true; |
| 2114 | } |
| 2115 | } |
| 2116 | Origin |= SymbolOrigin::AST; |
| 2117 | } |
| 2118 | if (Candidate.IdentifierResult) { |
| 2119 | Quality.References = Candidate.IdentifierResult->References; |
| 2120 | Relevance.Scope = SymbolRelevanceSignals::FileScope; |
| 2121 | Origin |= SymbolOrigin::Identifier; |
| 2122 | } |
| 2123 | } |
| 2124 | |
| 2125 | CodeCompletion::Scores Scores = evaluateCompletion(Quality, Relevance); |
| 2126 | if (Opts.RecordCCResult) |
| 2127 | Opts.RecordCCResult(toCodeCompletion(Bundle), Quality, Relevance, |
| 2128 | Scores.Total); |
| 2129 | |
| 2130 | dlog("CodeComplete: {0} ({1}) = {2}\n{3}{4}\n" , First.Name, |
| 2131 | llvm::to_string(Origin), Scores.Total, llvm::to_string(Quality), |
| 2132 | llvm::to_string(Relevance)); |
| 2133 | |
| 2134 | NSema += bool(Origin & SymbolOrigin::AST); |
| 2135 | NIndex += FromIndex; |
| 2136 | NSemaAndIndex += bool(Origin & SymbolOrigin::AST) && FromIndex; |
| 2137 | NIdent += bool(Origin & SymbolOrigin::Identifier); |
| 2138 | if (Candidates.push(V: {std::move(Bundle), Scores})) |
| 2139 | Incomplete = true; |
| 2140 | } |
| 2141 | |
| 2142 | CodeCompletion toCodeCompletion(const CompletionCandidate::Bundle &Bundle) { |
| 2143 | std::optional<CodeCompletionBuilder> Builder; |
| 2144 | for (const auto &Item : Bundle) { |
| 2145 | CodeCompletionString *SemaCCS = |
| 2146 | Item.SemaResult ? Recorder->codeCompletionString(R: *Item.SemaResult) |
| 2147 | : nullptr; |
| 2148 | if (!Builder) |
| 2149 | Builder.emplace(args: Recorder ? &Recorder->CCSema->getASTContext() : nullptr, |
| 2150 | args: Item, args&: SemaCCS, args&: AccessibleScopes, args&: *Inserter, args&: FileName, |
| 2151 | args&: CCContextKind, args: Opts, args&: IsUsingDeclaration, args&: NextTokenKind); |
| 2152 | else |
| 2153 | Builder->add(C: Item, SemaCCS, ContextKind: CCContextKind); |
| 2154 | } |
| 2155 | return Builder->build(); |
| 2156 | } |
| 2157 | }; |
| 2158 | |
| 2159 | } // namespace |
| 2160 | |
| 2161 | clang::CodeCompleteOptions CodeCompleteOptions::getClangCompleteOpts() const { |
| 2162 | clang::CodeCompleteOptions Result; |
| 2163 | Result.IncludeCodePatterns = |
| 2164 | EnableSnippets && (CodePatterns != Config::CodePatternsPolicy::None); |
| 2165 | Result.IncludeMacros = true; |
| 2166 | Result.IncludeGlobals = true; |
| 2167 | // We choose to include full comments and not do doxygen parsing in |
| 2168 | // completion. |
| 2169 | // FIXME: ideally, we should support doxygen in some form, e.g. do markdown |
| 2170 | // formatting of the comments. |
| 2171 | Result.IncludeBriefComments = false; |
| 2172 | |
| 2173 | // When an is used, Sema is responsible for completing the main file, |
| 2174 | // the index can provide results from the preamble. |
| 2175 | // Tell Sema not to deserialize the preamble to look for results. |
| 2176 | Result.LoadExternal = ForceLoadPreamble || !Index; |
| 2177 | Result.IncludeFixIts = IncludeFixIts; |
| 2178 | |
| 2179 | return Result; |
| 2180 | } |
| 2181 | |
| 2182 | CompletionPrefix guessCompletionPrefix(llvm::StringRef Content, |
| 2183 | unsigned Offset) { |
| 2184 | assert(Offset <= Content.size()); |
| 2185 | StringRef Rest = Content.take_front(N: Offset); |
| 2186 | CompletionPrefix Result; |
| 2187 | |
| 2188 | // Consume the unqualified name. We only handle ASCII characters. |
| 2189 | // isAsciiIdentifierContinue will let us match "0invalid", but we don't mind. |
| 2190 | while (!Rest.empty() && isAsciiIdentifierContinue(c: Rest.back())) |
| 2191 | Rest = Rest.drop_back(); |
| 2192 | Result.Name = Content.slice(Start: Rest.size(), End: Offset); |
| 2193 | |
| 2194 | // Consume qualifiers. |
| 2195 | while (Rest.consume_back(Suffix: "::" ) && !Rest.ends_with(Suffix: ":" )) // reject :::: |
| 2196 | while (!Rest.empty() && isAsciiIdentifierContinue(c: Rest.back())) |
| 2197 | Rest = Rest.drop_back(); |
| 2198 | Result.Qualifier = |
| 2199 | Content.slice(Start: Rest.size(), End: Result.Name.begin() - Content.begin()); |
| 2200 | |
| 2201 | return Result; |
| 2202 | } |
| 2203 | |
| 2204 | // Code complete the argument name on "/*" inside function call. |
| 2205 | // Offset should be pointing to the start of the comment, i.e.: |
| 2206 | // foo(^/*, rather than foo(/*^) where the cursor probably is. |
| 2207 | CodeCompleteResult (PathRef FileName, unsigned Offset, |
| 2208 | llvm::StringRef Prefix, |
| 2209 | const PreambleData *Preamble, |
| 2210 | const ParseInputs &ParseInput) { |
| 2211 | if (Preamble == nullptr) // Can't run without Sema. |
| 2212 | return CodeCompleteResult(); |
| 2213 | |
| 2214 | clang::CodeCompleteOptions Options; |
| 2215 | Options.IncludeGlobals = false; |
| 2216 | Options.IncludeMacros = false; |
| 2217 | Options.IncludeCodePatterns = false; |
| 2218 | Options.IncludeBriefComments = false; |
| 2219 | std::set<std::string> ParamNames; |
| 2220 | // We want to see signatures coming from newly introduced includes, hence a |
| 2221 | // full patch. |
| 2222 | semaCodeComplete( |
| 2223 | Consumer: std::make_unique<ParamNameCollector>(args&: Options, args&: ParamNames), Options, |
| 2224 | Input: {.FileName: FileName, .Offset: Offset, .Preamble: *Preamble, |
| 2225 | .Patch: PreamblePatch::createFullPatch(FileName, Modified: ParseInput, Baseline: *Preamble), |
| 2226 | .ParseInput: ParseInput}); |
| 2227 | if (ParamNames.empty()) |
| 2228 | return CodeCompleteResult(); |
| 2229 | |
| 2230 | CodeCompleteResult Result; |
| 2231 | Range CompletionRange; |
| 2232 | // Skip /* |
| 2233 | Offset += 2; |
| 2234 | CompletionRange.start = offsetToPosition(Code: ParseInput.Contents, Offset); |
| 2235 | CompletionRange.end = |
| 2236 | offsetToPosition(Code: ParseInput.Contents, Offset: Offset + Prefix.size()); |
| 2237 | Result.CompletionRange = CompletionRange; |
| 2238 | Result.Context = CodeCompletionContext::CCC_NaturalLanguage; |
| 2239 | for (llvm::StringRef Name : ParamNames) { |
| 2240 | if (!Name.starts_with(Prefix)) |
| 2241 | continue; |
| 2242 | CodeCompletion Item; |
| 2243 | Item.Name = Name.str() + "=*/" ; |
| 2244 | Item.FilterText = Item.Name; |
| 2245 | Item.Kind = CompletionItemKind::Text; |
| 2246 | Item.CompletionTokenRange = CompletionRange; |
| 2247 | Item.Origin = SymbolOrigin::AST; |
| 2248 | Result.Completions.push_back(x: Item); |
| 2249 | } |
| 2250 | |
| 2251 | return Result; |
| 2252 | } |
| 2253 | |
| 2254 | // If Offset is inside what looks like argument comment (e.g. |
| 2255 | // "/*^" or "/* foo^"), returns new offset pointing to the start of the /* |
| 2256 | // (place where semaCodeComplete should run). |
| 2257 | std::optional<unsigned> |
| 2258 | (llvm::StringRef Content) { |
| 2259 | while (!Content.empty() && isAsciiIdentifierContinue(c: Content.back())) |
| 2260 | Content = Content.drop_back(); |
| 2261 | Content = Content.rtrim(); |
| 2262 | if (Content.ends_with(Suffix: "/*" )) |
| 2263 | return Content.size() - 2; |
| 2264 | return std::nullopt; |
| 2265 | } |
| 2266 | |
| 2267 | CodeCompleteResult codeComplete(PathRef FileName, Position Pos, |
| 2268 | const PreambleData *Preamble, |
| 2269 | const ParseInputs &ParseInput, |
| 2270 | CodeCompleteOptions Opts, |
| 2271 | SpeculativeFuzzyFind *SpecFuzzyFind) { |
| 2272 | auto Offset = positionToOffset(Code: ParseInput.Contents, P: Pos); |
| 2273 | if (!Offset) { |
| 2274 | elog(Fmt: "Code completion position was invalid {0}" , Vals: Offset.takeError()); |
| 2275 | return CodeCompleteResult(); |
| 2276 | } |
| 2277 | |
| 2278 | auto Content = llvm::StringRef(ParseInput.Contents).take_front(N: *Offset); |
| 2279 | if (auto = maybeFunctionArgumentCommentStart(Content)) { |
| 2280 | // We are doing code completion of a comment, where we currently only |
| 2281 | // support completing param names in function calls. To do this, we |
| 2282 | // require information from Sema, but Sema's comment completion stops at |
| 2283 | // parsing, so we must move back the position before running it, extract |
| 2284 | // information we need and construct completion items ourselves. |
| 2285 | auto = Content.substr(Start: *OffsetBeforeComment + 2).trim(); |
| 2286 | return codeCompleteComment(FileName, Offset: *OffsetBeforeComment, Prefix: CommentPrefix, |
| 2287 | Preamble, ParseInput); |
| 2288 | } |
| 2289 | |
| 2290 | auto Flow = CodeCompleteFlow( |
| 2291 | FileName, Preamble ? Preamble->Includes : IncludeStructure(), |
| 2292 | SpecFuzzyFind, Opts); |
| 2293 | return (!Preamble || Opts.RunParser == CodeCompleteOptions::NeverParse) |
| 2294 | ? std::move(Flow).runWithoutSema(Content: ParseInput.Contents, Offset: *Offset, |
| 2295 | TFS: *ParseInput.TFS) |
| 2296 | : std::move(Flow).run(SemaCCInput: {.FileName: FileName, .Offset: *Offset, .Preamble: *Preamble, |
| 2297 | /*PreamblePatch=*/ |
| 2298 | .Patch: PreamblePatch::createMacroPatch( |
| 2299 | FileName, Modified: ParseInput, Baseline: *Preamble), |
| 2300 | .ParseInput: ParseInput}); |
| 2301 | } |
| 2302 | |
| 2303 | SignatureHelp signatureHelp(PathRef FileName, Position Pos, |
| 2304 | const PreambleData &Preamble, |
| 2305 | const ParseInputs &ParseInput, |
| 2306 | MarkupKind DocumentationFormat) { |
| 2307 | auto Offset = positionToOffset(Code: ParseInput.Contents, P: Pos); |
| 2308 | if (!Offset) { |
| 2309 | elog(Fmt: "Signature help position was invalid {0}" , Vals: Offset.takeError()); |
| 2310 | return SignatureHelp(); |
| 2311 | } |
| 2312 | SignatureHelp Result; |
| 2313 | clang::CodeCompleteOptions Options; |
| 2314 | Options.IncludeGlobals = false; |
| 2315 | Options.IncludeMacros = false; |
| 2316 | Options.IncludeCodePatterns = false; |
| 2317 | Options.IncludeBriefComments = false; |
| 2318 | semaCodeComplete( |
| 2319 | Consumer: std::make_unique<SignatureHelpCollector>(args&: Options, args&: DocumentationFormat, |
| 2320 | args: ParseInput.Index, args&: Result), |
| 2321 | Options, |
| 2322 | Input: {.FileName: FileName, .Offset: *Offset, .Preamble: Preamble, |
| 2323 | .Patch: PreamblePatch::createFullPatch(FileName, Modified: ParseInput, Baseline: Preamble), |
| 2324 | .ParseInput: ParseInput}); |
| 2325 | return Result; |
| 2326 | } |
| 2327 | |
| 2328 | bool isIndexedForCodeCompletion(const NamedDecl &ND, ASTContext &ASTCtx) { |
| 2329 | auto InTopLevelScope = [](const NamedDecl &ND) { |
| 2330 | switch (ND.getDeclContext()->getDeclKind()) { |
| 2331 | case Decl::TranslationUnit: |
| 2332 | case Decl::Namespace: |
| 2333 | case Decl::LinkageSpec: |
| 2334 | return true; |
| 2335 | default: |
| 2336 | break; |
| 2337 | }; |
| 2338 | return false; |
| 2339 | }; |
| 2340 | auto InClassScope = [](const NamedDecl &ND) { |
| 2341 | return ND.getDeclContext()->getDeclKind() == Decl::CXXRecord; |
| 2342 | }; |
| 2343 | // We only complete symbol's name, which is the same as the name of the |
| 2344 | // *primary* template in case of template specializations. |
| 2345 | if (isExplicitTemplateSpecialization(D: &ND)) |
| 2346 | return false; |
| 2347 | |
| 2348 | // Category decls are not useful on their own outside the interface or |
| 2349 | // implementation blocks. Moreover, sema already provides completion for |
| 2350 | // these, even if it requires preamble deserialization. So by excluding them |
| 2351 | // from the index, we reduce the noise in all the other completion scopes. |
| 2352 | if (llvm::isa<ObjCCategoryDecl>(Val: &ND) || llvm::isa<ObjCCategoryImplDecl>(Val: &ND)) |
| 2353 | return false; |
| 2354 | |
| 2355 | if (InTopLevelScope(ND)) |
| 2356 | return true; |
| 2357 | |
| 2358 | // Always index enum constants, even if they're not in the top level scope: |
| 2359 | // when |
| 2360 | // --all-scopes-completion is set, we'll want to complete those as well. |
| 2361 | if (const auto *EnumDecl = dyn_cast<clang::EnumDecl>(ND.getDeclContext())) |
| 2362 | return (InTopLevelScope(*EnumDecl) || InClassScope(*EnumDecl)); |
| 2363 | |
| 2364 | return false; |
| 2365 | } |
| 2366 | |
| 2367 | CompletionItem CodeCompletion::render(const CodeCompleteOptions &Opts) const { |
| 2368 | CompletionItem LSP; |
| 2369 | const auto *InsertInclude = Includes.empty() ? nullptr : &Includes[0]; |
| 2370 | // We could move our indicators from label into labelDetails->description. |
| 2371 | // In VSCode there are rendering issues that prevent these being aligned. |
| 2372 | LSP.label = ((InsertInclude && InsertInclude->Insertion) |
| 2373 | ? Opts.IncludeIndicator.Insert |
| 2374 | : Opts.IncludeIndicator.NoInsert) + |
| 2375 | (Opts.ShowOrigins ? "[" + llvm::to_string(Value: Origin) + "]" : "" ) + |
| 2376 | RequiredQualifier + Name; |
| 2377 | LSP.labelDetails.emplace(); |
| 2378 | LSP.labelDetails->detail = Signature; |
| 2379 | |
| 2380 | LSP.kind = Kind; |
| 2381 | LSP.detail = BundleSize > 1 |
| 2382 | ? std::string(llvm::formatv(Fmt: "[{0} overloads]" , Vals: BundleSize)) |
| 2383 | : ReturnType; |
| 2384 | LSP.deprecated = Deprecated; |
| 2385 | // Combine header information and documentation in LSP `documentation` field. |
| 2386 | // This is not quite right semantically, but tends to display well in editors. |
| 2387 | if (InsertInclude || Documentation) { |
| 2388 | markup::Document Doc; |
| 2389 | if (InsertInclude) |
| 2390 | Doc.addParagraph().appendText(Text: "From " ).appendCode(Code: InsertInclude->Header); |
| 2391 | if (Documentation) |
| 2392 | Doc.append(Other: *Documentation); |
| 2393 | LSP.documentation = renderDoc(Doc, Kind: Opts.DocumentationFormat); |
| 2394 | } |
| 2395 | LSP.sortText = sortText(Score: Score.Total, Tiebreak: FilterText); |
| 2396 | LSP.filterText = FilterText; |
| 2397 | LSP.textEdit = {.range: CompletionTokenRange, .newText: RequiredQualifier + Name, .annotationId: "" }; |
| 2398 | // Merge continuous additionalTextEdits into main edit. The main motivation |
| 2399 | // behind this is to help LSP clients, it seems most of them are confused when |
| 2400 | // they are provided with additionalTextEdits that are consecutive to main |
| 2401 | // edit. |
| 2402 | // Note that we store additional text edits from back to front in a line. That |
| 2403 | // is mainly to help LSP clients again, so that changes do not effect each |
| 2404 | // other. |
| 2405 | for (const auto &FixIt : FixIts) { |
| 2406 | if (FixIt.range.end == LSP.textEdit->range.start) { |
| 2407 | LSP.textEdit->newText = FixIt.newText + LSP.textEdit->newText; |
| 2408 | LSP.textEdit->range.start = FixIt.range.start; |
| 2409 | } else { |
| 2410 | LSP.additionalTextEdits.push_back(x: FixIt); |
| 2411 | } |
| 2412 | } |
| 2413 | if (Opts.EnableSnippets) |
| 2414 | LSP.textEdit->newText += SnippetSuffix; |
| 2415 | |
| 2416 | // FIXME(kadircet): Do not even fill insertText after making sure textEdit is |
| 2417 | // compatible with most of the editors. |
| 2418 | LSP.insertText = LSP.textEdit->newText; |
| 2419 | // Some clients support snippets but work better with plaintext. |
| 2420 | // So if the snippet is trivial, let the client know. |
| 2421 | // https://github.com/clangd/clangd/issues/922 |
| 2422 | LSP.insertTextFormat = (Opts.EnableSnippets && !SnippetSuffix.empty()) |
| 2423 | ? InsertTextFormat::Snippet |
| 2424 | : InsertTextFormat::PlainText; |
| 2425 | if (InsertInclude && InsertInclude->Insertion) |
| 2426 | LSP.additionalTextEdits.push_back(x: *InsertInclude->Insertion); |
| 2427 | |
| 2428 | LSP.score = Score.ExcludingName; |
| 2429 | |
| 2430 | return LSP; |
| 2431 | } |
| 2432 | |
| 2433 | llvm::raw_ostream &operator<<(llvm::raw_ostream &OS, const CodeCompletion &C) { |
| 2434 | // For now just lean on CompletionItem. |
| 2435 | return OS << C.render(Opts: CodeCompleteOptions()); |
| 2436 | } |
| 2437 | |
| 2438 | llvm::raw_ostream &operator<<(llvm::raw_ostream &OS, |
| 2439 | const CodeCompleteResult &R) { |
| 2440 | OS << "CodeCompleteResult: " << R.Completions.size() << (R.HasMore ? "+" : "" ) |
| 2441 | << " (" << getCompletionKindString(Kind: R.Context) << ")" |
| 2442 | << " items:\n" ; |
| 2443 | for (const auto &C : R.Completions) |
| 2444 | OS << C << "\n" ; |
| 2445 | return OS; |
| 2446 | } |
| 2447 | |
| 2448 | // Heuristically detect whether the `Line` is an unterminated include filename. |
| 2449 | bool isIncludeFile(llvm::StringRef Line) { |
| 2450 | Line = Line.ltrim(); |
| 2451 | if (!Line.consume_front(Prefix: "#" )) |
| 2452 | return false; |
| 2453 | Line = Line.ltrim(); |
| 2454 | if (!(Line.consume_front(Prefix: "include_next" ) || Line.consume_front(Prefix: "include" ) || |
| 2455 | Line.consume_front(Prefix: "import" ))) |
| 2456 | return false; |
| 2457 | Line = Line.ltrim(); |
| 2458 | if (Line.consume_front(Prefix: "<" )) |
| 2459 | return Line.count(C: '>') == 0; |
| 2460 | if (Line.consume_front(Prefix: "\"" )) |
| 2461 | return Line.count(C: '"') == 0; |
| 2462 | return false; |
| 2463 | } |
| 2464 | |
| 2465 | bool allowImplicitCompletion(llvm::StringRef Content, unsigned Offset) { |
| 2466 | // Look at last line before completion point only. |
| 2467 | Content = Content.take_front(N: Offset); |
| 2468 | auto Pos = Content.rfind(C: '\n'); |
| 2469 | if (Pos != llvm::StringRef::npos) |
| 2470 | Content = Content.substr(Start: Pos + 1); |
| 2471 | |
| 2472 | // Complete after scope operators. |
| 2473 | if (Content.ends_with(Suffix: "." ) || Content.ends_with(Suffix: "->" ) || |
| 2474 | Content.ends_with(Suffix: "::" ) || Content.ends_with(Suffix: "/*" )) |
| 2475 | return true; |
| 2476 | // Complete after `#include <` and #include `<foo/`. |
| 2477 | if ((Content.ends_with(Suffix: "<" ) || Content.ends_with(Suffix: "\"" ) || |
| 2478 | Content.ends_with(Suffix: "/" )) && |
| 2479 | isIncludeFile(Line: Content)) |
| 2480 | return true; |
| 2481 | |
| 2482 | // Complete words. Give non-ascii characters the benefit of the doubt. |
| 2483 | return !Content.empty() && (isAsciiIdentifierContinue(c: Content.back()) || |
| 2484 | !llvm::isASCII(C: Content.back())); |
| 2485 | } |
| 2486 | |
| 2487 | } // namespace clangd |
| 2488 | } // namespace clang |
| 2489 | |