| 1 | //===--- ExtractVariable.cpp ------------------------------------*- C++-*-===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | #include "AST.h" |
| 9 | #include "ParsedAST.h" |
| 10 | #include "Protocol.h" |
| 11 | #include "Selection.h" |
| 12 | #include "SourceCode.h" |
| 13 | #include "refactor/Tweak.h" |
| 14 | #include "clang/AST/ASTContext.h" |
| 15 | #include "clang/AST/Decl.h" |
| 16 | #include "clang/AST/DeclCXX.h" |
| 17 | #include "clang/AST/Expr.h" |
| 18 | #include "clang/AST/ExprCXX.h" |
| 19 | #include "clang/AST/LambdaCapture.h" |
| 20 | #include "clang/AST/OperationKinds.h" |
| 21 | #include "clang/AST/RecursiveASTVisitor.h" |
| 22 | #include "clang/AST/Stmt.h" |
| 23 | #include "clang/AST/StmtCXX.h" |
| 24 | #include "clang/Basic/LangOptions.h" |
| 25 | #include "clang/Basic/SourceLocation.h" |
| 26 | #include "clang/Basic/SourceManager.h" |
| 27 | #include "clang/Tooling/Core/Replacement.h" |
| 28 | #include "llvm/ADT/SmallVector.h" |
| 29 | #include "llvm/ADT/StringRef.h" |
| 30 | #include "llvm/Support/Casting.h" |
| 31 | #include "llvm/Support/Error.h" |
| 32 | #include "llvm/Support/raw_ostream.h" |
| 33 | |
| 34 | namespace clang { |
| 35 | namespace clangd { |
| 36 | namespace { |
| 37 | // information regarding the Expr that is being extracted |
| 38 | class { |
| 39 | public: |
| 40 | ExtractionContext(const SelectionTree::Node *Node, const SourceManager &SM, |
| 41 | const ASTContext &Ctx); |
| 42 | const clang::Expr *() const { return Expr; } |
| 43 | const SelectionTree::Node *() const { return ExprNode; } |
| 44 | bool () const { return Extractable; } |
| 45 | // The half-open range for the expression to be extracted. |
| 46 | SourceRange getExtractionChars() const; |
| 47 | // Generate Replacement for replacing selected expression with given VarName |
| 48 | tooling::Replacement replaceWithVar(SourceRange Chars, |
| 49 | llvm::StringRef VarName) const; |
| 50 | // Generate Replacement for declaring the selected Expr as a new variable |
| 51 | tooling::Replacement insertDeclaration(llvm::StringRef VarName, |
| 52 | SourceRange InitChars, |
| 53 | bool AddSemicolon) const; |
| 54 | |
| 55 | private: |
| 56 | bool = false; |
| 57 | const clang::Expr *; |
| 58 | QualType ; |
| 59 | const SelectionTree::Node *; |
| 60 | // Stmt before which we will extract |
| 61 | const clang::Stmt * = nullptr; |
| 62 | const SourceManager &; |
| 63 | const ASTContext &; |
| 64 | // Decls referenced in the Expr |
| 65 | std::vector<clang::Decl *> ; |
| 66 | // returns true if the Expr doesn't reference any variable declared in scope |
| 67 | bool exprIsValidOutside(const clang::Stmt *Scope) const; |
| 68 | // computes the Stmt before which we will extract out Expr |
| 69 | const clang::Stmt *computeInsertionPoint() const; |
| 70 | }; |
| 71 | |
| 72 | // Returns all the Decls referenced inside the given Expr |
| 73 | static std::vector<clang::Decl *> |
| 74 | computeReferencedDecls(const clang::Expr *Expr) { |
| 75 | // RAV subclass to find all DeclRefs in a given Stmt |
| 76 | class FindDeclRefsVisitor |
| 77 | : public clang::RecursiveASTVisitor<FindDeclRefsVisitor> { |
| 78 | public: |
| 79 | std::vector<Decl *> ReferencedDecls; |
| 80 | bool VisitDeclRefExpr(DeclRefExpr *DeclRef) { // NOLINT |
| 81 | // Stop the call operator of lambdas from being marked as a referenced |
| 82 | // DeclRefExpr in immediately invoked lambdas. |
| 83 | if (const auto *const Method = |
| 84 | llvm::dyn_cast<CXXMethodDecl>(Val: DeclRef->getDecl()); |
| 85 | Method != nullptr && Method->getParent()->isLambda()) { |
| 86 | return true; |
| 87 | } |
| 88 | ReferencedDecls.push_back(DeclRef->getDecl()); |
| 89 | return true; |
| 90 | } |
| 91 | |
| 92 | // Local variables declared inside of the selected lambda cannot go out of |
| 93 | // scope. The DeclRefExprs that are important are the variables captured, |
| 94 | // the DeclRefExprs inside the initializers of init-capture variables, |
| 95 | // variables mentioned in trailing return types, constraints and explicit |
| 96 | // defaulted template parameters. |
| 97 | bool TraverseLambdaExpr(LambdaExpr *LExpr) { |
| 98 | for (const auto &[Capture, Initializer] : |
| 99 | llvm::zip(t: LExpr->captures(), u: LExpr->capture_inits())) { |
| 100 | TraverseLambdaCapture(LE: LExpr, C: &Capture, Init: Initializer); |
| 101 | } |
| 102 | |
| 103 | if (const clang::Expr *RequiresClause = |
| 104 | LExpr->getTrailingRequiresClause().ConstraintExpr) { |
| 105 | TraverseStmt(const_cast<clang::Expr *>(RequiresClause)); |
| 106 | } |
| 107 | |
| 108 | for (auto *const TemplateParam : LExpr->getExplicitTemplateParameters()) |
| 109 | TraverseDecl(TemplateParam); |
| 110 | |
| 111 | if (auto *const CallOperator = LExpr->getCallOperator()) { |
| 112 | TraverseType(T: CallOperator->getDeclaredReturnType()); |
| 113 | |
| 114 | for (auto *const Param : CallOperator->parameters()) { |
| 115 | TraverseParmVarDecl(Param); |
| 116 | } |
| 117 | |
| 118 | for (auto *const Attr : CallOperator->attrs()) { |
| 119 | TraverseAttr(Attr); |
| 120 | } |
| 121 | } |
| 122 | |
| 123 | return true; |
| 124 | } |
| 125 | }; |
| 126 | |
| 127 | FindDeclRefsVisitor Visitor; |
| 128 | Visitor.TraverseStmt(S: const_cast<Stmt *>(cast<Stmt>(Val: Expr))); |
| 129 | return Visitor.ReferencedDecls; |
| 130 | } |
| 131 | |
| 132 | static QualType computeVariableType(const Expr *Expr, const ASTContext &Ctx) { |
| 133 | if (Ctx.getLangOpts().CPlusPlus11) |
| 134 | return Ctx.getAutoDeductType(); |
| 135 | |
| 136 | if (Expr->hasPlaceholderType(K: BuiltinType::PseudoObject)) { |
| 137 | if (const auto *PR = dyn_cast<ObjCPropertyRefExpr>(Val: Expr)) { |
| 138 | if (PR->isMessagingSetter()) { |
| 139 | // Don't support extracting a compound reference like `self.prop += 1` |
| 140 | // since the meaning changes after extraction since we'll no longer call |
| 141 | // the setter. Non compound access like `self.prop = 1` is invalid since |
| 142 | // it returns nil (setter method must have a void return type). |
| 143 | return QualType(); |
| 144 | } else if (PR->isMessagingGetter()) { |
| 145 | if (PR->isExplicitProperty()) |
| 146 | return PR->getExplicitProperty()->getType(); |
| 147 | else |
| 148 | return PR->getImplicitPropertyGetter()->getReturnType(); |
| 149 | } |
| 150 | } else { |
| 151 | return QualType(); |
| 152 | } |
| 153 | } |
| 154 | return Expr->getType(); |
| 155 | } |
| 156 | |
| 157 | ExtractionContext::(const SelectionTree::Node *Node, |
| 158 | const SourceManager &SM, |
| 159 | const ASTContext &Ctx) |
| 160 | : ExprNode(Node), SM(SM), Ctx(Ctx) { |
| 161 | Expr = Node->ASTNode.get<clang::Expr>(); |
| 162 | ReferencedDecls = computeReferencedDecls(Expr); |
| 163 | InsertionPoint = computeInsertionPoint(); |
| 164 | if (InsertionPoint) |
| 165 | Extractable = true; |
| 166 | VarType = computeVariableType(Expr, Ctx); |
| 167 | if (VarType.isNull()) |
| 168 | Extractable = false; |
| 169 | else |
| 170 | // Strip the outer nullability since it's not common for local variables. |
| 171 | AttributedType::stripOuterNullability(VarType); |
| 172 | } |
| 173 | |
| 174 | // checks whether extracting before InsertionPoint will take a |
| 175 | // variable reference out of scope |
| 176 | bool ExtractionContext::(const clang::Stmt *Scope) const { |
| 177 | SourceLocation ScopeBegin = Scope->getBeginLoc(); |
| 178 | SourceLocation ScopeEnd = Scope->getEndLoc(); |
| 179 | for (const Decl *ReferencedDecl : ReferencedDecls) { |
| 180 | if (ReferencedDecl->getBeginLoc().isValid() && |
| 181 | SM.isPointWithin(Location: ReferencedDecl->getBeginLoc(), Start: ScopeBegin, End: ScopeEnd) && |
| 182 | SM.isPointWithin(Location: ReferencedDecl->getEndLoc(), Start: ScopeBegin, End: ScopeEnd)) |
| 183 | return false; |
| 184 | } |
| 185 | return true; |
| 186 | } |
| 187 | |
| 188 | // Return the Stmt before which we need to insert the extraction. |
| 189 | // To find the Stmt, we go up the AST Tree and if the Parent of the current |
| 190 | // Stmt is a CompoundStmt, we can extract inside this CompoundStmt just before |
| 191 | // the current Stmt. We ALWAYS insert before a Stmt whose parent is a |
| 192 | // CompoundStmt |
| 193 | // |
| 194 | // FIXME: Extraction from label, switch and case statements |
| 195 | // FIXME: Doens't work for FoldExpr |
| 196 | // FIXME: Ensure extraction from loops doesn't change semantics. |
| 197 | const clang::Stmt *ExtractionContext::() const { |
| 198 | // returns true if we can extract before InsertionPoint |
| 199 | auto = |
| 200 | [](const SelectionTree::Node *InsertionPoint) -> bool { |
| 201 | if (const clang::Stmt *Stmt = InsertionPoint->ASTNode.get<clang::Stmt>()) { |
| 202 | if (isa<clang::Expr>(Val: Stmt)) { |
| 203 | // Do not allow extraction from the initializer of a defaulted parameter |
| 204 | // to a local variable (e.g. a function-local lambda). |
| 205 | if (InsertionPoint->Parent->ASTNode.get<ParmVarDecl>() != nullptr) { |
| 206 | return false; |
| 207 | } |
| 208 | |
| 209 | return true; |
| 210 | } |
| 211 | |
| 212 | // We don't yet allow extraction from switch/case stmt as we would need to |
| 213 | // jump over the switch stmt even if there is a CompoundStmt inside the |
| 214 | // switch. And there are other Stmts which we don't care about (e.g. |
| 215 | // continue and break) as there can never be anything to extract from |
| 216 | // them. |
| 217 | return isa<AttributedStmt>(Val: Stmt) || isa<CompoundStmt>(Val: Stmt) || |
| 218 | isa<CXXForRangeStmt>(Val: Stmt) || isa<DeclStmt>(Val: Stmt) || |
| 219 | isa<DoStmt>(Val: Stmt) || isa<ForStmt>(Val: Stmt) || isa<IfStmt>(Val: Stmt) || |
| 220 | isa<ReturnStmt>(Val: Stmt) || isa<WhileStmt>(Val: Stmt); |
| 221 | } |
| 222 | if (InsertionPoint->ASTNode.get<VarDecl>()) |
| 223 | return true; |
| 224 | return false; |
| 225 | }; |
| 226 | for (const SelectionTree::Node *CurNode = getExprNode(); |
| 227 | CurNode->Parent && CanExtractOutside(CurNode); |
| 228 | CurNode = CurNode->Parent) { |
| 229 | const clang::Stmt *CurInsertionPoint = CurNode->ASTNode.get<Stmt>(); |
| 230 | // give up if extraction will take a variable out of scope |
| 231 | if (CurInsertionPoint && !exprIsValidOutside(Scope: CurInsertionPoint)) |
| 232 | break; |
| 233 | if (const clang::Stmt *CurParent = CurNode->Parent->ASTNode.get<Stmt>()) { |
| 234 | if (isa<CompoundStmt>(Val: CurParent)) { |
| 235 | // Ensure we don't write inside a macro. |
| 236 | if (CurParent->getBeginLoc().isMacroID()) |
| 237 | continue; |
| 238 | return CurInsertionPoint; |
| 239 | } |
| 240 | } |
| 241 | } |
| 242 | return nullptr; |
| 243 | } |
| 244 | |
| 245 | // returns the replacement for substituting the extraction with VarName |
| 246 | tooling::Replacement |
| 247 | ExtractionContext::(SourceRange Chars, |
| 248 | llvm::StringRef VarName) const { |
| 249 | unsigned = |
| 250 | SM.getFileOffset(SpellingLoc: Chars.getEnd()) - SM.getFileOffset(SpellingLoc: Chars.getBegin()); |
| 251 | return tooling::Replacement(SM, Chars.getBegin(), ExtractionLength, VarName); |
| 252 | } |
| 253 | // returns the Replacement for declaring a new variable storing the extraction |
| 254 | tooling::Replacement |
| 255 | ExtractionContext::(llvm::StringRef VarName, |
| 256 | SourceRange InitializerChars, |
| 257 | bool AddSemicolon) const { |
| 258 | llvm::StringRef = toSourceCode(SM, R: InitializerChars); |
| 259 | const SourceLocation InsertionLoc = |
| 260 | toHalfOpenFileRange(Mgr: SM, LangOpts: Ctx.getLangOpts(), |
| 261 | R: InsertionPoint->getSourceRange()) |
| 262 | ->getBegin(); |
| 263 | std::string = |
| 264 | printType(VarType, ExprNode->getDeclContext(), VarName) + " = " + |
| 265 | ExtractionCode.str(); |
| 266 | if (AddSemicolon) |
| 267 | ExtractedVarDecl += "; " ; |
| 268 | return tooling::Replacement(SM, InsertionLoc, 0, ExtractedVarDecl); |
| 269 | } |
| 270 | |
| 271 | // Helpers for handling "binary subexpressions" like a + [[b + c]] + d. |
| 272 | // |
| 273 | // These are special, because the formal AST doesn't match what users expect: |
| 274 | // - the AST is ((a + b) + c) + d, so the ancestor expression is `a + b + c`. |
| 275 | // - but extracting `b + c` is reasonable, as + is (mathematically) associative. |
| 276 | // |
| 277 | // So we try to support these cases with some restrictions: |
| 278 | // - the operator must be associative |
| 279 | // - no mixing of operators is allowed |
| 280 | // - we don't look inside macro expansions in the subexpressions |
| 281 | // - we only adjust the extracted range, so references in the unselected parts |
| 282 | // of the AST expression (e.g. `a`) are still considered referenced for |
| 283 | // the purposes of calculating the insertion point. |
| 284 | // FIXME: it would be nice to exclude these references, by micromanaging |
| 285 | // the computeReferencedDecls() calls around the binary operator tree. |
| 286 | |
| 287 | // Information extracted about a binary operator encounted in a SelectionTree. |
| 288 | // It can represent either an overloaded or built-in operator. |
| 289 | struct ParsedBinaryOperator { |
| 290 | BinaryOperatorKind Kind; |
| 291 | SourceLocation ExprLoc; |
| 292 | llvm::SmallVector<const SelectionTree::Node *> SelectedOperands; |
| 293 | |
| 294 | // If N is a binary operator, populate this and return true. |
| 295 | bool parse(const SelectionTree::Node &N) { |
| 296 | SelectedOperands.clear(); |
| 297 | |
| 298 | if (const BinaryOperator *Op = |
| 299 | llvm::dyn_cast_or_null<BinaryOperator>(N.ASTNode.get<Expr>())) { |
| 300 | Kind = Op->getOpcode(); |
| 301 | ExprLoc = Op->getExprLoc(); |
| 302 | SelectedOperands = N.Children; |
| 303 | return true; |
| 304 | } |
| 305 | if (const CXXOperatorCallExpr *Op = |
| 306 | llvm::dyn_cast_or_null<CXXOperatorCallExpr>( |
| 307 | N.ASTNode.get<Expr>())) { |
| 308 | if (!Op->isInfixBinaryOp()) |
| 309 | return false; |
| 310 | |
| 311 | Kind = BinaryOperator::getOverloadedOpcode(OO: Op->getOperator()); |
| 312 | ExprLoc = Op->getExprLoc(); |
| 313 | // Not all children are args, there's also the callee (operator). |
| 314 | for (const auto *Child : N.Children) { |
| 315 | const Expr *E = Child->ASTNode.get<Expr>(); |
| 316 | assert(E && "callee and args should be Exprs!" ); |
| 317 | if (E == Op->getArg(0) || E == Op->getArg(1)) |
| 318 | SelectedOperands.push_back(Elt: Child); |
| 319 | } |
| 320 | return true; |
| 321 | } |
| 322 | return false; |
| 323 | } |
| 324 | |
| 325 | bool associative() const { |
| 326 | // Must also be left-associative, or update getBinaryOperatorRange()! |
| 327 | switch (Kind) { |
| 328 | case BO_Add: |
| 329 | case BO_Mul: |
| 330 | case BO_And: |
| 331 | case BO_Or: |
| 332 | case BO_Xor: |
| 333 | case BO_LAnd: |
| 334 | case BO_LOr: |
| 335 | return true; |
| 336 | default: |
| 337 | return false; |
| 338 | } |
| 339 | } |
| 340 | |
| 341 | bool crossesMacroBoundary(const SourceManager &SM) { |
| 342 | FileID F = SM.getFileID(SpellingLoc: ExprLoc); |
| 343 | for (const SelectionTree::Node *Child : SelectedOperands) |
| 344 | if (SM.getFileID(Child->ASTNode.get<Expr>()->getExprLoc()) != F) |
| 345 | return true; |
| 346 | return false; |
| 347 | } |
| 348 | }; |
| 349 | |
| 350 | // If have an associative operator at the top level, then we must find |
| 351 | // the start point (rightmost in LHS) and end point (leftmost in RHS). |
| 352 | // We can only descend into subtrees where the operator matches. |
| 353 | // |
| 354 | // e.g. for a + [[b + c]] + d |
| 355 | // + |
| 356 | // / \ |
| 357 | // N-> + d |
| 358 | // / \ |
| 359 | // + c <- End |
| 360 | // / \ |
| 361 | // a b <- Start |
| 362 | const SourceRange getBinaryOperatorRange(const SelectionTree::Node &N, |
| 363 | const SourceManager &SM, |
| 364 | const LangOptions &LangOpts) { |
| 365 | // If N is not a suitable binary operator, bail out. |
| 366 | ParsedBinaryOperator Op; |
| 367 | if (!Op.parse(N: N.ignoreImplicit()) || !Op.associative() || |
| 368 | Op.crossesMacroBoundary(SM) || Op.SelectedOperands.size() != 2) |
| 369 | return SourceRange(); |
| 370 | BinaryOperatorKind OuterOp = Op.Kind; |
| 371 | |
| 372 | // Because the tree we're interested in contains only one operator type, and |
| 373 | // all eligible operators are left-associative, the shape of the tree is |
| 374 | // very restricted: it's a linked list along the left edges. |
| 375 | // This simplifies our implementation. |
| 376 | const SelectionTree::Node *Start = Op.SelectedOperands.front(); // LHS |
| 377 | const SelectionTree::Node *End = Op.SelectedOperands.back(); // RHS |
| 378 | // End is already correct: it can't be an OuterOp (as it's left-associative). |
| 379 | // Start needs to be pushed down int the subtree to the right spot. |
| 380 | while (Op.parse(N: Start->ignoreImplicit()) && Op.Kind == OuterOp && |
| 381 | !Op.crossesMacroBoundary(SM)) { |
| 382 | assert(!Op.SelectedOperands.empty() && "got only operator on one side!" ); |
| 383 | if (Op.SelectedOperands.size() == 1) { // Only Op.RHS selected |
| 384 | Start = Op.SelectedOperands.back(); |
| 385 | break; |
| 386 | } |
| 387 | // Op.LHS is (at least partially) selected, so descend into it. |
| 388 | Start = Op.SelectedOperands.front(); |
| 389 | } |
| 390 | |
| 391 | return SourceRange( |
| 392 | toHalfOpenFileRange(SM, LangOpts, Start->ASTNode.getSourceRange()) |
| 393 | ->getBegin(), |
| 394 | toHalfOpenFileRange(SM, LangOpts, End->ASTNode.getSourceRange()) |
| 395 | ->getEnd()); |
| 396 | } |
| 397 | |
| 398 | SourceRange ExtractionContext::() const { |
| 399 | // Special case: we're extracting an associative binary subexpression. |
| 400 | SourceRange BinaryOperatorRange = |
| 401 | getBinaryOperatorRange(N: *ExprNode, SM, LangOpts: Ctx.getLangOpts()); |
| 402 | if (BinaryOperatorRange.isValid()) |
| 403 | return BinaryOperatorRange; |
| 404 | |
| 405 | // Usual case: we're extracting the whole expression. |
| 406 | return *toHalfOpenFileRange(SM, Ctx.getLangOpts(), Expr->getSourceRange()); |
| 407 | } |
| 408 | |
| 409 | // Find the CallExpr whose callee is the (possibly wrapped) DeclRef |
| 410 | const SelectionTree::Node *getCallExpr(const SelectionTree::Node *DeclRef) { |
| 411 | const SelectionTree::Node &MaybeCallee = DeclRef->outerImplicit(); |
| 412 | const SelectionTree::Node *MaybeCall = MaybeCallee.Parent; |
| 413 | if (!MaybeCall) |
| 414 | return nullptr; |
| 415 | const CallExpr *CE = |
| 416 | llvm::dyn_cast_or_null<CallExpr>(MaybeCall->ASTNode.get<Expr>()); |
| 417 | if (!CE) |
| 418 | return nullptr; |
| 419 | if (CE->getCallee() != MaybeCallee.ASTNode.get<Expr>()) |
| 420 | return nullptr; |
| 421 | return MaybeCall; |
| 422 | } |
| 423 | |
| 424 | // Returns true if Inner (which is a direct child of Outer) is appearing as |
| 425 | // a statement rather than an expression whose value can be used. |
| 426 | bool childExprIsDisallowedStmt(const Stmt *Outer, const Expr *Inner) { |
| 427 | if (!Outer || !Inner) |
| 428 | return false; |
| 429 | // Exclude the most common places where an expr can appear but be unused. |
| 430 | if (llvm::isa<SwitchCase>(Val: Outer)) |
| 431 | return true; |
| 432 | // Control flow statements use condition etc, but not the body. |
| 433 | if (const auto *WS = llvm::dyn_cast<WhileStmt>(Val: Outer)) |
| 434 | return Inner == WS->getBody(); |
| 435 | if (const auto *DS = llvm::dyn_cast<DoStmt>(Val: Outer)) |
| 436 | return Inner == DS->getBody(); |
| 437 | if (const auto *FS = llvm::dyn_cast<ForStmt>(Val: Outer)) |
| 438 | return Inner == FS->getBody(); |
| 439 | if (const auto *FS = llvm::dyn_cast<CXXForRangeStmt>(Val: Outer)) |
| 440 | return Inner == FS->getBody(); |
| 441 | if (const auto *IS = llvm::dyn_cast<IfStmt>(Val: Outer)) |
| 442 | return Inner == IS->getThen() || Inner == IS->getElse(); |
| 443 | // Assume all other cases may be actual expressions. |
| 444 | // This includes the important case of subexpressions (where Outer is Expr). |
| 445 | return false; |
| 446 | } |
| 447 | |
| 448 | // check if N can and should be extracted (e.g. is not void-typed). |
| 449 | bool (const SelectionTree::Node *N) { |
| 450 | const Expr *E = N->ASTNode.get<Expr>(); |
| 451 | if (!E) |
| 452 | return false; |
| 453 | |
| 454 | // Void expressions can't be assigned to variables. |
| 455 | const Type *ExprType = E->getType().getTypePtrOrNull(); |
| 456 | if (!ExprType || ExprType->isVoidType()) |
| 457 | return false; |
| 458 | |
| 459 | // A plain reference to a name (e.g. variable) isn't worth extracting. |
| 460 | // FIXME: really? What if it's e.g. `std::is_same<void, void>::value`? |
| 461 | if (llvm::isa<DeclRefExpr>(Val: E)) |
| 462 | return false; |
| 463 | |
| 464 | // Similarly disallow extraction for member exprs with an implicit `this`. |
| 465 | if (const auto *ME = dyn_cast<MemberExpr>(E)) |
| 466 | if (const auto *TE = dyn_cast<CXXThisExpr>(ME->getBase()->IgnoreImpCasts())) |
| 467 | if (TE->isImplicit()) |
| 468 | return false; |
| 469 | |
| 470 | // Extracting Exprs like a = 1 gives placeholder = a = 1 which isn't useful. |
| 471 | // FIXME: we could still hoist the assignment, and leave the variable there? |
| 472 | ParsedBinaryOperator BinOp; |
| 473 | bool IsBinOp = BinOp.parse(N: *N); |
| 474 | if (IsBinOp && BinaryOperator::isAssignmentOp(Opc: BinOp.Kind)) |
| 475 | return false; |
| 476 | |
| 477 | const SelectionTree::Node &OuterImplicit = N->outerImplicit(); |
| 478 | const auto *Parent = OuterImplicit.Parent; |
| 479 | if (!Parent) |
| 480 | return false; |
| 481 | // Filter non-applicable expression statements. |
| 482 | if (childExprIsDisallowedStmt(Parent->ASTNode.get<Stmt>(), |
| 483 | OuterImplicit.ASTNode.get<Expr>())) |
| 484 | return false; |
| 485 | |
| 486 | std::function<bool(const SelectionTree::Node *)> IsFullySelected = |
| 487 | [&](const SelectionTree::Node *N) { |
| 488 | if (N->ASTNode.getSourceRange().isValid() && |
| 489 | N->Selected != SelectionTree::Complete) |
| 490 | return false; |
| 491 | for (const auto *Child : N->Children) { |
| 492 | if (!IsFullySelected(Child)) |
| 493 | return false; |
| 494 | } |
| 495 | return true; |
| 496 | }; |
| 497 | auto ExprIsFullySelectedTargetNode = [&](const Expr *E) { |
| 498 | if (E != OuterImplicit.ASTNode.get<Expr>()) |
| 499 | return false; |
| 500 | |
| 501 | // The above condition is the only relevant one except for binary operators. |
| 502 | // Without the following code, we would fail to offer extraction for e.g.: |
| 503 | // int x = 1 + 2 + [[3 + 4 + 5]]; |
| 504 | // See the documentation of ParsedBinaryOperator for further details. |
| 505 | if (!IsBinOp) |
| 506 | return true; |
| 507 | return IsFullySelected(N); |
| 508 | }; |
| 509 | |
| 510 | // Disable extraction of full RHS on assignment operations, e.g: |
| 511 | // x = [[RHS_EXPR]]; |
| 512 | // This would just result in duplicating the code. |
| 513 | if (const auto *BO = Parent->ASTNode.get<BinaryOperator>()) { |
| 514 | if (BO->isAssignmentOp() && ExprIsFullySelectedTargetNode(BO->getRHS())) |
| 515 | return false; |
| 516 | } |
| 517 | |
| 518 | // If e.g. a capture clause was selected, the target node is the lambda |
| 519 | // expression. We only want to offer the extraction if the entire lambda |
| 520 | // expression was selected. |
| 521 | if (llvm::isa<LambdaExpr>(Val: E)) |
| 522 | return N->Selected == SelectionTree::Complete; |
| 523 | |
| 524 | // The same logic as for assignments applies to initializations. |
| 525 | // However, we do allow extracting the RHS of an init capture, as it is |
| 526 | // a valid use case to move non-trivial expressions out of the capture clause. |
| 527 | // FIXME: In that case, the extracted variable should be captured directly, |
| 528 | // rather than an explicit copy. |
| 529 | if (const auto *Decl = Parent->ASTNode.get<VarDecl>()) { |
| 530 | if (!Decl->isInitCapture() && |
| 531 | ExprIsFullySelectedTargetNode(Decl->getInit())) { |
| 532 | return false; |
| 533 | } |
| 534 | } |
| 535 | |
| 536 | return true; |
| 537 | } |
| 538 | |
| 539 | // Find the Expr node that we're going to extract. |
| 540 | // We don't want to trigger for assignment expressions and variable/field |
| 541 | // DeclRefs. For function/member function, we want to extract the entire |
| 542 | // function call. |
| 543 | const SelectionTree::Node *(const SelectionTree::Node *N) { |
| 544 | if (!N) |
| 545 | return nullptr; |
| 546 | const SelectionTree::Node *TargetNode = N; |
| 547 | const clang::Expr *SelectedExpr = N->ASTNode.get<clang::Expr>(); |
| 548 | if (!SelectedExpr) |
| 549 | return nullptr; |
| 550 | // For function and member function DeclRefs, extract the whole call. |
| 551 | if (llvm::isa<DeclRefExpr>(Val: SelectedExpr) || |
| 552 | llvm::isa<MemberExpr>(Val: SelectedExpr)) |
| 553 | if (const SelectionTree::Node *Call = getCallExpr(DeclRef: N)) |
| 554 | TargetNode = Call; |
| 555 | // Extracting Exprs like a = 1 gives placeholder = a = 1 which isn't useful. |
| 556 | if (const BinaryOperator *BinOpExpr = |
| 557 | dyn_cast_or_null<BinaryOperator>(Val: SelectedExpr)) { |
| 558 | if (BinOpExpr->getOpcode() == BinaryOperatorKind::BO_Assign) |
| 559 | return nullptr; |
| 560 | } |
| 561 | if (!TargetNode || !eligibleForExtraction(N: TargetNode)) |
| 562 | return nullptr; |
| 563 | return TargetNode; |
| 564 | } |
| 565 | |
| 566 | /// Extracts an expression to the variable placeholder |
| 567 | /// Before: |
| 568 | /// int x = 5 + 4 * 3; |
| 569 | /// ^^^^^ |
| 570 | /// After: |
| 571 | /// auto placeholder = 5 + 4; |
| 572 | /// int x = placeholder * 3; |
| 573 | class : public Tweak { |
| 574 | public: |
| 575 | const char *id() const final; |
| 576 | bool prepare(const Selection &Inputs) override; |
| 577 | Expected<Effect> apply(const Selection &Inputs) override; |
| 578 | std::string () const override { |
| 579 | return "Extract subexpression to variable" ; |
| 580 | } |
| 581 | llvm::StringLiteral () const override { |
| 582 | return CodeAction::REFACTOR_KIND; |
| 583 | } |
| 584 | |
| 585 | private: |
| 586 | // the expression to extract |
| 587 | std::unique_ptr<ExtractionContext> ; |
| 588 | }; |
| 589 | REGISTER_TWEAK(ExtractVariable) |
| 590 | bool ExtractVariable::(const Selection &Inputs) { |
| 591 | // we don't trigger on empty selections for now |
| 592 | if (Inputs.SelectionBegin == Inputs.SelectionEnd) |
| 593 | return false; |
| 594 | const ASTContext &Ctx = Inputs.AST->getASTContext(); |
| 595 | const SourceManager &SM = Inputs.AST->getSourceManager(); |
| 596 | if (const SelectionTree::Node *N = |
| 597 | computeExtractedExpr(N: Inputs.ASTSelection.commonAncestor())) |
| 598 | Target = std::make_unique<ExtractionContext>(args&: N, args: SM, args: Ctx); |
| 599 | return Target && Target->isExtractable(); |
| 600 | } |
| 601 | |
| 602 | Expected<Tweak::Effect> ExtractVariable::(const Selection &Inputs) { |
| 603 | tooling::Replacements Result; |
| 604 | // FIXME: get variable name from user or suggest based on type |
| 605 | std::string VarName = "placeholder" ; |
| 606 | SourceRange Range = Target->getExtractionChars(); |
| 607 | |
| 608 | const SelectionTree::Node &OuterImplicit = |
| 609 | Target->getExprNode()->outerImplicit(); |
| 610 | assert(OuterImplicit.Parent); |
| 611 | bool IsExprStmt = llvm::isa_and_nonnull<CompoundStmt>( |
| 612 | OuterImplicit.Parent->ASTNode.get<Stmt>()); |
| 613 | |
| 614 | // insert new variable declaration. add a semicolon if and only if |
| 615 | // we are not dealing with an expression statement, which already has |
| 616 | // a semicolon that stays where it is, as it's not part of the range. |
| 617 | if (auto Err = |
| 618 | Result.add(Target->insertDeclaration(VarName, Range, !IsExprStmt))) |
| 619 | return std::move(Err); |
| 620 | |
| 621 | // replace expression with variable name, unless it's an expression statement, |
| 622 | // in which case we remove it. |
| 623 | if (IsExprStmt) |
| 624 | VarName.clear(); |
| 625 | if (auto Err = Result.add(R: Target->replaceWithVar(Chars: Range, VarName))) |
| 626 | return std::move(Err); |
| 627 | return Effect::mainFileEdit(SM: Inputs.AST->getSourceManager(), |
| 628 | Replacements: std::move(Result)); |
| 629 | } |
| 630 | |
| 631 | } // namespace |
| 632 | } // namespace clangd |
| 633 | } // namespace clang |
| 634 | |