Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 1 | //===--- ExtractFunction.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 | // Extracts statements to a new function and replaces the statements with a |
| 10 | // call to the new function. |
| 11 | // Before: |
| 12 | // void f(int a) { |
| 13 | // [[if(a < 5) |
| 14 | // a = 5;]] |
| 15 | // } |
| 16 | // After: |
| 17 | // void extracted(int &a) { |
| 18 | // if(a < 5) |
| 19 | // a = 5; |
| 20 | // } |
| 21 | // void f(int a) { |
| 22 | // extracted(a); |
| 23 | // } |
| 24 | // |
| 25 | // - Only extract statements |
| 26 | // - Extracts from non-templated free functions only. |
| 27 | // - Parameters are const only if the declaration was const |
| 28 | // - Always passed by l-value reference |
| 29 | // - Void return type |
| 30 | // - Cannot extract declarations that will be needed in the original function |
| 31 | // after extraction. |
Shaurya Gupta | 3b08a61f | 2019-08-30 09:57:56 +0000 | [diff] [blame] | 32 | // - Checks for broken control flow (break/continue without loop/switch) |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 33 | // |
| 34 | // 1. ExtractFunction is the tweak subclass |
| 35 | // - Prepare does basic analysis of the selection and is therefore fast. |
| 36 | // Successful prepare doesn't always mean we can apply the tweak. |
| 37 | // - Apply does a more detailed analysis and can be slower. In case of |
| 38 | // failure, we let the user know that we are unable to perform extraction. |
| 39 | // 2. ExtractionZone store information about the range being extracted and the |
| 40 | // enclosing function. |
| 41 | // 3. NewFunction stores properties of the extracted function and provides |
| 42 | // methods for rendering it. |
| 43 | // 4. CapturedZoneInfo uses a RecursiveASTVisitor to capture information about |
| 44 | // the extraction like declarations, existing return statements, etc. |
| 45 | // 5. getExtractedFunction is responsible for analyzing the CapturedZoneInfo and |
| 46 | // creating a NewFunction. |
| 47 | //===----------------------------------------------------------------------===// |
| 48 | |
| 49 | #include "AST.h" |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 50 | #include "Logger.h" |
Sam McCall | 915f978 | 2019-09-04 09:46:06 +0000 | [diff] [blame] | 51 | #include "ParsedAST.h" |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 52 | #include "Selection.h" |
| 53 | #include "SourceCode.h" |
| 54 | #include "refactor/Tweak.h" |
| 55 | #include "clang/AST/ASTContext.h" |
| 56 | #include "clang/AST/Decl.h" |
| 57 | #include "clang/AST/DeclTemplate.h" |
| 58 | #include "clang/AST/RecursiveASTVisitor.h" |
| 59 | #include "clang/AST/Stmt.h" |
| 60 | #include "clang/Basic/LangOptions.h" |
| 61 | #include "clang/Basic/SourceLocation.h" |
| 62 | #include "clang/Basic/SourceManager.h" |
| 63 | #include "clang/Lex/Lexer.h" |
| 64 | #include "clang/Tooling/Core/Replacement.h" |
| 65 | #include "clang/Tooling/Refactoring/Extract/SourceExtraction.h" |
| 66 | #include "llvm/ADT/None.h" |
| 67 | #include "llvm/ADT/Optional.h" |
| 68 | #include "llvm/ADT/SmallVector.h" |
| 69 | #include "llvm/ADT/StringRef.h" |
| 70 | #include "llvm/ADT/iterator_range.h" |
| 71 | #include "llvm/Support/Casting.h" |
| 72 | #include "llvm/Support/Error.h" |
| 73 | |
| 74 | namespace clang { |
| 75 | namespace clangd { |
| 76 | namespace { |
| 77 | |
| 78 | using Node = SelectionTree::Node; |
| 79 | |
| 80 | // ExtractionZone is the part of code that is being extracted. |
| 81 | // EnclosingFunction is the function/method inside which the zone lies. |
| 82 | // We split the file into 4 parts relative to extraction zone. |
| 83 | enum class ZoneRelative { |
| 84 | Before, // Before Zone and inside EnclosingFunction. |
| 85 | Inside, // Inside Zone. |
| 86 | After, // After Zone and inside EnclosingFunction. |
| 87 | OutsideFunc // Outside EnclosingFunction. |
| 88 | }; |
| 89 | |
| 90 | // A RootStmt is a statement that's fully selected including all it's children |
| 91 | // and it's parent is unselected. |
| 92 | // Check if a node is a root statement. |
| 93 | bool isRootStmt(const Node *N) { |
| 94 | if (!N->ASTNode.get<Stmt>()) |
| 95 | return false; |
| 96 | // Root statement cannot be partially selected. |
| 97 | if (N->Selected == SelectionTree::Partial) |
| 98 | return false; |
| 99 | // Only DeclStmt can be an unselected RootStmt since VarDecls claim the entire |
| 100 | // selection range in selectionTree. |
| 101 | if (N->Selected == SelectionTree::Unselected && !N->ASTNode.get<DeclStmt>()) |
| 102 | return false; |
| 103 | return true; |
| 104 | } |
| 105 | |
| 106 | // Returns the (unselected) parent of all RootStmts given the commonAncestor. |
| 107 | // We only support extraction of RootStmts since it allows us to extract without |
| 108 | // having to change the selection range. Also, this means that any scope that |
| 109 | // begins in selection range, ends in selection range and any scope that begins |
| 110 | // outside the selection range, ends outside as well. |
| 111 | const Node *getParentOfRootStmts(const Node *CommonAnc) { |
| 112 | if (!CommonAnc) |
| 113 | return nullptr; |
| 114 | switch (CommonAnc->Selected) { |
| 115 | case SelectionTree::Selection::Unselected: |
| 116 | // Ensure all Children are RootStmts. |
| 117 | return llvm::all_of(CommonAnc->Children, isRootStmt) ? CommonAnc : nullptr; |
| 118 | case SelectionTree::Selection::Partial: |
| 119 | // Treat Partially selected VarDecl as completely selected since |
| 120 | // SelectionTree doesn't always select VarDecls correctly. |
| 121 | // FIXME: Remove this after D66872 is upstream) |
| 122 | if (!CommonAnc->ASTNode.get<VarDecl>()) |
| 123 | return nullptr; |
| 124 | LLVM_FALLTHROUGH; |
| 125 | case SelectionTree::Selection::Complete: |
| 126 | // If the Common Ancestor is completely selected, then it's a root statement |
| 127 | // and its parent will be unselected. |
| 128 | const Node *Parent = CommonAnc->Parent; |
| 129 | // If parent is a DeclStmt, even though it's unselected, we consider it a |
| 130 | // root statement and return its parent. This is done because the VarDecls |
| 131 | // claim the entire selection range of the Declaration and DeclStmt is |
| 132 | // always unselected. |
| 133 | return Parent->ASTNode.get<DeclStmt>() ? Parent->Parent : Parent; |
| 134 | } |
Simon Pilgrim | 46b346e | 2019-08-29 12:37:02 +0000 | [diff] [blame] | 135 | llvm_unreachable("Unhandled SelectionTree::Selection enum"); |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 136 | } |
| 137 | |
| 138 | // The ExtractionZone class forms a view of the code wrt Zone. |
| 139 | struct ExtractionZone { |
| 140 | // Parent of RootStatements being extracted. |
| 141 | const Node *Parent = nullptr; |
| 142 | // The half-open file range of the code being extracted. |
| 143 | SourceRange ZoneRange; |
| 144 | // The function inside which our zone resides. |
| 145 | const FunctionDecl *EnclosingFunction = nullptr; |
| 146 | // The half-open file range of the enclosing function. |
| 147 | SourceRange EnclosingFuncRange; |
| 148 | SourceLocation getInsertionPoint() const { |
| 149 | return EnclosingFuncRange.getBegin(); |
| 150 | } |
| 151 | bool isRootStmt(const Stmt *S) const; |
| 152 | // The last root statement is important to decide where we need to insert a |
| 153 | // semicolon after the extraction. |
| 154 | const Node *getLastRootStmt() const { return Parent->Children.back(); } |
| 155 | void generateRootStmts(); |
Shaurya Gupta | 3b08a61f | 2019-08-30 09:57:56 +0000 | [diff] [blame] | 156 | |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 157 | private: |
| 158 | llvm::DenseSet<const Stmt *> RootStmts; |
| 159 | }; |
| 160 | |
| 161 | bool ExtractionZone::isRootStmt(const Stmt *S) const { |
| 162 | return RootStmts.find(S) != RootStmts.end(); |
| 163 | } |
| 164 | |
| 165 | // Generate RootStmts set |
| 166 | void ExtractionZone::generateRootStmts() { |
Shaurya Gupta | 3b08a61f | 2019-08-30 09:57:56 +0000 | [diff] [blame] | 167 | for (const Node *Child : Parent->Children) |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 168 | RootStmts.insert(Child->ASTNode.get<Stmt>()); |
| 169 | } |
| 170 | |
| 171 | // Finds the function in which the zone lies. |
| 172 | const FunctionDecl *findEnclosingFunction(const Node *CommonAnc) { |
| 173 | // Walk up the SelectionTree until we find a function Decl |
| 174 | for (const Node *CurNode = CommonAnc; CurNode; CurNode = CurNode->Parent) { |
| 175 | // Don't extract from lambdas |
| 176 | if (CurNode->ASTNode.get<LambdaExpr>()) |
| 177 | return nullptr; |
| 178 | if (const FunctionDecl *Func = CurNode->ASTNode.get<FunctionDecl>()) { |
| 179 | // FIXME: Support extraction from methods. |
| 180 | if (isa<CXXMethodDecl>(Func)) |
| 181 | return nullptr; |
| 182 | // FIXME: Support extraction from templated functions. |
Shaurya Gupta | 3b08a61f | 2019-08-30 09:57:56 +0000 | [diff] [blame] | 183 | if (Func->isTemplated()) |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 184 | return nullptr; |
| 185 | return Func; |
| 186 | } |
| 187 | } |
| 188 | return nullptr; |
| 189 | } |
| 190 | |
| 191 | // Zone Range is the union of SourceRanges of all child Nodes in Parent since |
| 192 | // all child Nodes are RootStmts |
| 193 | llvm::Optional<SourceRange> findZoneRange(const Node *Parent, |
| 194 | const SourceManager &SM, |
| 195 | const LangOptions &LangOpts) { |
| 196 | SourceRange SR; |
| 197 | if (auto BeginFileRange = toHalfOpenFileRange( |
| 198 | SM, LangOpts, Parent->Children.front()->ASTNode.getSourceRange())) |
| 199 | SR.setBegin(BeginFileRange->getBegin()); |
| 200 | else |
| 201 | return llvm::None; |
| 202 | if (auto EndFileRange = toHalfOpenFileRange( |
| 203 | SM, LangOpts, Parent->Children.back()->ASTNode.getSourceRange())) |
| 204 | SR.setEnd(EndFileRange->getEnd()); |
| 205 | else |
| 206 | return llvm::None; |
| 207 | return SR; |
| 208 | } |
| 209 | |
| 210 | // Compute the range spanned by the enclosing function. |
| 211 | // FIXME: check if EnclosingFunction has any attributes as the AST doesn't |
| 212 | // always store the source range of the attributes and thus we end up extracting |
| 213 | // between the attributes and the EnclosingFunction. |
| 214 | llvm::Optional<SourceRange> |
| 215 | computeEnclosingFuncRange(const FunctionDecl *EnclosingFunction, |
| 216 | const SourceManager &SM, |
| 217 | const LangOptions &LangOpts) { |
| 218 | return toHalfOpenFileRange(SM, LangOpts, EnclosingFunction->getSourceRange()); |
| 219 | } |
| 220 | |
| 221 | // FIXME: Check we're not extracting from the initializer/condition of a control |
| 222 | // flow structure. |
| 223 | // FIXME: Check that we don't extract the compound statement of the |
| 224 | // enclosingFunction. |
| 225 | llvm::Optional<ExtractionZone> findExtractionZone(const Node *CommonAnc, |
| 226 | const SourceManager &SM, |
| 227 | const LangOptions &LangOpts) { |
| 228 | ExtractionZone ExtZone; |
| 229 | ExtZone.Parent = getParentOfRootStmts(CommonAnc); |
| 230 | if (!ExtZone.Parent || ExtZone.Parent->Children.empty()) |
| 231 | return llvm::None; |
| 232 | // Don't extract expressions. |
| 233 | // FIXME: We should extract expressions that are "statements" i.e. not |
| 234 | // subexpressions |
| 235 | if (ExtZone.Parent->Children.size() == 1 && |
| 236 | ExtZone.getLastRootStmt()->ASTNode.get<Expr>()) |
| 237 | return llvm::None; |
| 238 | ExtZone.EnclosingFunction = findEnclosingFunction(ExtZone.Parent); |
| 239 | if (!ExtZone.EnclosingFunction) |
| 240 | return llvm::None; |
| 241 | if (auto FuncRange = |
| 242 | computeEnclosingFuncRange(ExtZone.EnclosingFunction, SM, LangOpts)) |
| 243 | ExtZone.EnclosingFuncRange = *FuncRange; |
| 244 | if (auto ZoneRange = findZoneRange(ExtZone.Parent, SM, LangOpts)) |
| 245 | ExtZone.ZoneRange = *ZoneRange; |
| 246 | if (ExtZone.EnclosingFuncRange.isInvalid() || ExtZone.ZoneRange.isInvalid()) |
| 247 | return llvm::None; |
| 248 | ExtZone.generateRootStmts(); |
| 249 | return ExtZone; |
| 250 | } |
| 251 | |
| 252 | // Stores information about the extracted function and provides methods for |
| 253 | // rendering it. |
| 254 | struct NewFunction { |
| 255 | struct Parameter { |
| 256 | std::string Name; |
| 257 | QualType TypeInfo; |
| 258 | bool PassByReference; |
| 259 | unsigned OrderPriority; // Lower value parameters are preferred first. |
| 260 | std::string render(const DeclContext *Context) const; |
| 261 | bool operator<(const Parameter &Other) const { |
| 262 | return OrderPriority < Other.OrderPriority; |
| 263 | } |
| 264 | }; |
| 265 | std::string Name = "extracted"; |
| 266 | std::string ReturnType; |
| 267 | std::vector<Parameter> Parameters; |
| 268 | SourceRange BodyRange; |
| 269 | SourceLocation InsertionPoint; |
| 270 | const DeclContext *EnclosingFuncContext; |
| 271 | // Decides whether the extracted function body and the function call need a |
| 272 | // semicolon after extraction. |
| 273 | tooling::ExtractionSemicolonPolicy SemicolonPolicy; |
| 274 | NewFunction(tooling::ExtractionSemicolonPolicy SemicolonPolicy) |
| 275 | : SemicolonPolicy(SemicolonPolicy) {} |
| 276 | // Render the call for this function. |
| 277 | std::string renderCall() const; |
| 278 | // Render the definition for this function. |
| 279 | std::string renderDefinition(const SourceManager &SM) const; |
| 280 | |
| 281 | private: |
| 282 | std::string renderParametersForDefinition() const; |
| 283 | std::string renderParametersForCall() const; |
| 284 | // Generate the function body. |
| 285 | std::string getFuncBody(const SourceManager &SM) const; |
| 286 | }; |
| 287 | |
| 288 | std::string NewFunction::renderParametersForDefinition() const { |
| 289 | std::string Result; |
| 290 | bool NeedCommaBefore = false; |
| 291 | for (const Parameter &P : Parameters) { |
| 292 | if (NeedCommaBefore) |
| 293 | Result += ", "; |
| 294 | NeedCommaBefore = true; |
| 295 | Result += P.render(EnclosingFuncContext); |
| 296 | } |
| 297 | return Result; |
| 298 | } |
| 299 | |
| 300 | std::string NewFunction::renderParametersForCall() const { |
| 301 | std::string Result; |
| 302 | bool NeedCommaBefore = false; |
| 303 | for (const Parameter &P : Parameters) { |
| 304 | if (NeedCommaBefore) |
| 305 | Result += ", "; |
| 306 | NeedCommaBefore = true; |
| 307 | Result += P.Name; |
| 308 | } |
| 309 | return Result; |
| 310 | } |
| 311 | |
| 312 | std::string NewFunction::renderCall() const { |
| 313 | return Name + "(" + renderParametersForCall() + ")" + |
| 314 | (SemicolonPolicy.isNeededInOriginalFunction() ? ";" : ""); |
| 315 | } |
| 316 | |
| 317 | std::string NewFunction::renderDefinition(const SourceManager &SM) const { |
| 318 | return ReturnType + " " + Name + "(" + renderParametersForDefinition() + ")" + |
| 319 | " {\n" + getFuncBody(SM) + "\n}\n"; |
| 320 | } |
| 321 | |
| 322 | std::string NewFunction::getFuncBody(const SourceManager &SM) const { |
| 323 | // FIXME: Generate tooling::Replacements instead of std::string to |
| 324 | // - hoist decls |
| 325 | // - add return statement |
| 326 | // - Add semicolon |
| 327 | return toSourceCode(SM, BodyRange).str() + |
| 328 | (SemicolonPolicy.isNeededInExtractedFunction() ? ";" : ""); |
| 329 | } |
| 330 | |
| 331 | std::string NewFunction::Parameter::render(const DeclContext *Context) const { |
| 332 | return printType(TypeInfo, *Context) + (PassByReference ? " &" : " ") + Name; |
| 333 | } |
| 334 | |
| 335 | // Stores captured information about Extraction Zone. |
| 336 | struct CapturedZoneInfo { |
| 337 | struct DeclInformation { |
| 338 | const Decl *TheDecl; |
| 339 | ZoneRelative DeclaredIn; |
| 340 | // index of the declaration or first reference. |
| 341 | unsigned DeclIndex; |
| 342 | bool IsReferencedInZone = false; |
| 343 | bool IsReferencedInPostZone = false; |
| 344 | // FIXME: Capture mutation information |
| 345 | DeclInformation(const Decl *TheDecl, ZoneRelative DeclaredIn, |
| 346 | unsigned DeclIndex) |
| 347 | : TheDecl(TheDecl), DeclaredIn(DeclaredIn), DeclIndex(DeclIndex){}; |
| 348 | // Marks the occurence of a reference for this declaration |
| 349 | void markOccurence(ZoneRelative ReferenceLoc); |
| 350 | }; |
| 351 | // Maps Decls to their DeclInfo |
| 352 | llvm::DenseMap<const Decl *, DeclInformation> DeclInfoMap; |
| 353 | // True if there is a return statement in zone. |
| 354 | bool HasReturnStmt = false; |
Shaurya Gupta | 3b08a61f | 2019-08-30 09:57:56 +0000 | [diff] [blame] | 355 | // Control flow is broken if we are extracting a break/continue without a |
| 356 | // corresponding parent loop/switch |
| 357 | bool BrokenControlFlow = false; |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 358 | // FIXME: capture TypeAliasDecl and UsingDirectiveDecl |
| 359 | // FIXME: Capture type information as well. |
| 360 | DeclInformation *createDeclInfo(const Decl *D, ZoneRelative RelativeLoc); |
| 361 | DeclInformation *getDeclInfoFor(const Decl *D); |
| 362 | }; |
| 363 | |
| 364 | CapturedZoneInfo::DeclInformation * |
| 365 | CapturedZoneInfo::createDeclInfo(const Decl *D, ZoneRelative RelativeLoc) { |
| 366 | // The new Decl's index is the size of the map so far. |
| 367 | auto InsertionResult = DeclInfoMap.insert( |
| 368 | {D, DeclInformation(D, RelativeLoc, DeclInfoMap.size())}); |
| 369 | // Return the newly created DeclInfo |
| 370 | return &InsertionResult.first->second; |
| 371 | } |
| 372 | |
| 373 | CapturedZoneInfo::DeclInformation * |
| 374 | CapturedZoneInfo::getDeclInfoFor(const Decl *D) { |
| 375 | // If the Decl doesn't exist, we |
| 376 | auto Iter = DeclInfoMap.find(D); |
| 377 | if (Iter == DeclInfoMap.end()) |
| 378 | return nullptr; |
| 379 | return &Iter->second; |
| 380 | } |
| 381 | |
| 382 | void CapturedZoneInfo::DeclInformation::markOccurence( |
| 383 | ZoneRelative ReferenceLoc) { |
| 384 | switch (ReferenceLoc) { |
| 385 | case ZoneRelative::Inside: |
| 386 | IsReferencedInZone = true; |
| 387 | break; |
| 388 | case ZoneRelative::After: |
| 389 | IsReferencedInPostZone = true; |
| 390 | break; |
| 391 | default: |
| 392 | break; |
| 393 | } |
| 394 | } |
| 395 | |
Shaurya Gupta | 3b08a61f | 2019-08-30 09:57:56 +0000 | [diff] [blame] | 396 | bool isLoop(const Stmt *S) { |
| 397 | return isa<ForStmt>(S) || isa<DoStmt>(S) || isa<WhileStmt>(S) || |
| 398 | isa<CXXForRangeStmt>(S); |
| 399 | } |
| 400 | |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 401 | // Captures information from Extraction Zone |
| 402 | CapturedZoneInfo captureZoneInfo(const ExtractionZone &ExtZone) { |
| 403 | // We use the ASTVisitor instead of using the selection tree since we need to |
| 404 | // find references in the PostZone as well. |
| 405 | // FIXME: Check which statements we don't allow to extract. |
| 406 | class ExtractionZoneVisitor |
| 407 | : public clang::RecursiveASTVisitor<ExtractionZoneVisitor> { |
| 408 | public: |
| 409 | ExtractionZoneVisitor(const ExtractionZone &ExtZone) : ExtZone(ExtZone) { |
| 410 | TraverseDecl(const_cast<FunctionDecl *>(ExtZone.EnclosingFunction)); |
| 411 | } |
Shaurya Gupta | 3b08a61f | 2019-08-30 09:57:56 +0000 | [diff] [blame] | 412 | |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 413 | bool TraverseStmt(Stmt *S) { |
Shaurya Gupta | 3b08a61f | 2019-08-30 09:57:56 +0000 | [diff] [blame] | 414 | if (!S) |
| 415 | return true; |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 416 | bool IsRootStmt = ExtZone.isRootStmt(const_cast<const Stmt *>(S)); |
| 417 | // If we are starting traversal of a RootStmt, we are somewhere inside |
| 418 | // ExtractionZone |
| 419 | if (IsRootStmt) |
| 420 | CurrentLocation = ZoneRelative::Inside; |
Shaurya Gupta | 3b08a61f | 2019-08-30 09:57:56 +0000 | [diff] [blame] | 421 | addToLoopSwitchCounters(S, 1); |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 422 | // Traverse using base class's TraverseStmt |
| 423 | RecursiveASTVisitor::TraverseStmt(S); |
Shaurya Gupta | 3b08a61f | 2019-08-30 09:57:56 +0000 | [diff] [blame] | 424 | addToLoopSwitchCounters(S, -1); |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 425 | // We set the current location as after since next stmt will either be a |
| 426 | // RootStmt (handled at the beginning) or after extractionZone |
| 427 | if (IsRootStmt) |
| 428 | CurrentLocation = ZoneRelative::After; |
| 429 | return true; |
| 430 | } |
Shaurya Gupta | 3b08a61f | 2019-08-30 09:57:56 +0000 | [diff] [blame] | 431 | |
| 432 | // Add Increment to CurNumberOf{Loops,Switch} if statement is |
| 433 | // {Loop,Switch} and inside Extraction Zone. |
| 434 | void addToLoopSwitchCounters(Stmt *S, int Increment) { |
| 435 | if (CurrentLocation != ZoneRelative::Inside) |
| 436 | return; |
| 437 | if (isLoop(S)) |
| 438 | CurNumberOfNestedLoops += Increment; |
| 439 | else if (isa<SwitchStmt>(S)) |
| 440 | CurNumberOfSwitch += Increment; |
| 441 | } |
| 442 | |
| 443 | // Decrement CurNumberOf{NestedLoops,Switch} if statement is {Loop,Switch} |
| 444 | // and inside Extraction Zone. |
| 445 | void decrementLoopSwitchCounters(Stmt *S) { |
| 446 | if (CurrentLocation != ZoneRelative::Inside) |
| 447 | return; |
| 448 | if (isLoop(S)) |
| 449 | CurNumberOfNestedLoops--; |
| 450 | else if (isa<SwitchStmt>(S)) |
| 451 | CurNumberOfSwitch--; |
| 452 | } |
| 453 | |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 454 | bool VisitDecl(Decl *D) { |
| 455 | Info.createDeclInfo(D, CurrentLocation); |
| 456 | return true; |
| 457 | } |
Shaurya Gupta | 3b08a61f | 2019-08-30 09:57:56 +0000 | [diff] [blame] | 458 | |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 459 | bool VisitDeclRefExpr(DeclRefExpr *DRE) { |
| 460 | // Find the corresponding Decl and mark it's occurence. |
| 461 | const Decl *D = DRE->getDecl(); |
| 462 | auto *DeclInfo = Info.getDeclInfoFor(D); |
| 463 | // If no Decl was found, the Decl must be outside the enclosingFunc. |
| 464 | if (!DeclInfo) |
| 465 | DeclInfo = Info.createDeclInfo(D, ZoneRelative::OutsideFunc); |
| 466 | DeclInfo->markOccurence(CurrentLocation); |
| 467 | // FIXME: check if reference mutates the Decl being referred. |
| 468 | return true; |
| 469 | } |
Shaurya Gupta | 3b08a61f | 2019-08-30 09:57:56 +0000 | [diff] [blame] | 470 | |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 471 | bool VisitReturnStmt(ReturnStmt *Return) { |
| 472 | if (CurrentLocation == ZoneRelative::Inside) |
| 473 | Info.HasReturnStmt = true; |
| 474 | return true; |
| 475 | } |
| 476 | |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 477 | bool VisitBreakStmt(BreakStmt *Break) { |
Shaurya Gupta | 3b08a61f | 2019-08-30 09:57:56 +0000 | [diff] [blame] | 478 | // Control flow is broken if break statement is selected without any |
| 479 | // parent loop or switch statement. |
| 480 | if (CurrentLocation == ZoneRelative::Inside && |
| 481 | !(CurNumberOfNestedLoops || CurNumberOfSwitch)) |
| 482 | Info.BrokenControlFlow = true; |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 483 | return true; |
| 484 | } |
Shaurya Gupta | 3b08a61f | 2019-08-30 09:57:56 +0000 | [diff] [blame] | 485 | |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 486 | bool VisitContinueStmt(ContinueStmt *Continue) { |
Shaurya Gupta | 3b08a61f | 2019-08-30 09:57:56 +0000 | [diff] [blame] | 487 | // Control flow is broken if Continue statement is selected without any |
| 488 | // parent loop |
| 489 | if (CurrentLocation == ZoneRelative::Inside && !CurNumberOfNestedLoops) |
| 490 | Info.BrokenControlFlow = true; |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 491 | return true; |
| 492 | } |
| 493 | CapturedZoneInfo Info; |
| 494 | const ExtractionZone &ExtZone; |
| 495 | ZoneRelative CurrentLocation = ZoneRelative::Before; |
Shaurya Gupta | 3b08a61f | 2019-08-30 09:57:56 +0000 | [diff] [blame] | 496 | // Number of {loop,switch} statements that are currently in the traversal |
| 497 | // stack inside Extraction Zone. Used to check for broken control flow. |
| 498 | unsigned CurNumberOfNestedLoops = 0; |
| 499 | unsigned CurNumberOfSwitch = 0; |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 500 | }; |
| 501 | ExtractionZoneVisitor Visitor(ExtZone); |
| 502 | return std::move(Visitor.Info); |
| 503 | } |
| 504 | |
| 505 | // Adds parameters to ExtractedFunc. |
| 506 | // Returns true if able to find the parameters successfully and no hoisting |
| 507 | // needed. |
Shaurya Gupta | 6a80145 | 2019-08-29 15:11:59 +0000 | [diff] [blame] | 508 | // FIXME: Check if the declaration has a local/anonymous type |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 509 | bool createParameters(NewFunction &ExtractedFunc, |
| 510 | const CapturedZoneInfo &CapturedInfo) { |
| 511 | for (const auto &KeyVal : CapturedInfo.DeclInfoMap) { |
| 512 | const auto &DeclInfo = KeyVal.second; |
| 513 | // If a Decl was Declared in zone and referenced in post zone, it |
| 514 | // needs to be hoisted (we bail out in that case). |
| 515 | // FIXME: Support Decl Hoisting. |
| 516 | if (DeclInfo.DeclaredIn == ZoneRelative::Inside && |
| 517 | DeclInfo.IsReferencedInPostZone) |
| 518 | return false; |
| 519 | if (!DeclInfo.IsReferencedInZone) |
| 520 | continue; // no need to pass as parameter, not referenced |
| 521 | if (DeclInfo.DeclaredIn == ZoneRelative::Inside || |
| 522 | DeclInfo.DeclaredIn == ZoneRelative::OutsideFunc) |
| 523 | continue; // no need to pass as parameter, still accessible. |
| 524 | // Parameter specific checks. |
| 525 | const ValueDecl *VD = dyn_cast_or_null<ValueDecl>(DeclInfo.TheDecl); |
| 526 | // Can't parameterise if the Decl isn't a ValueDecl or is a FunctionDecl |
| 527 | // (this includes the case of recursive call to EnclosingFunc in Zone). |
| 528 | if (!VD || isa<FunctionDecl>(DeclInfo.TheDecl)) |
| 529 | return false; |
| 530 | // Parameter qualifiers are same as the Decl's qualifiers. |
| 531 | QualType TypeInfo = VD->getType().getNonReferenceType(); |
| 532 | // FIXME: Need better qualifier checks: check mutated status for |
| 533 | // Decl(e.g. was it assigned, passed as nonconst argument, etc) |
| 534 | // FIXME: check if parameter will be a non l-value reference. |
| 535 | // FIXME: We don't want to always pass variables of types like int, |
| 536 | // pointers, etc by reference. |
| 537 | bool IsPassedByReference = true; |
| 538 | // We use the index of declaration as the ordering priority for parameters. |
| 539 | ExtractedFunc.Parameters.push_back( |
| 540 | {VD->getName(), TypeInfo, IsPassedByReference, DeclInfo.DeclIndex}); |
| 541 | } |
| 542 | llvm::sort(ExtractedFunc.Parameters); |
| 543 | return true; |
| 544 | } |
| 545 | |
| 546 | // Clangd uses open ranges while ExtractionSemicolonPolicy (in Clang Tooling) |
| 547 | // uses closed ranges. Generates the semicolon policy for the extraction and |
| 548 | // extends the ZoneRange if necessary. |
| 549 | tooling::ExtractionSemicolonPolicy |
| 550 | getSemicolonPolicy(ExtractionZone &ExtZone, const SourceManager &SM, |
| 551 | const LangOptions &LangOpts) { |
| 552 | // Get closed ZoneRange. |
| 553 | SourceRange FuncBodyRange = {ExtZone.ZoneRange.getBegin(), |
| 554 | ExtZone.ZoneRange.getEnd().getLocWithOffset(-1)}; |
| 555 | auto SemicolonPolicy = tooling::ExtractionSemicolonPolicy::compute( |
| 556 | ExtZone.getLastRootStmt()->ASTNode.get<Stmt>(), FuncBodyRange, SM, |
| 557 | LangOpts); |
| 558 | // Update ZoneRange. |
| 559 | ExtZone.ZoneRange.setEnd(FuncBodyRange.getEnd().getLocWithOffset(1)); |
| 560 | return SemicolonPolicy; |
| 561 | } |
| 562 | |
| 563 | // Generate return type for ExtractedFunc. Return false if unable to do so. |
| 564 | bool generateReturnProperties(NewFunction &ExtractedFunc, |
| 565 | const CapturedZoneInfo &CapturedInfo) { |
| 566 | |
| 567 | // FIXME: Use Existing Return statements (if present) |
| 568 | // FIXME: Generate new return statement if needed. |
| 569 | if (CapturedInfo.HasReturnStmt) |
| 570 | return false; |
| 571 | ExtractedFunc.ReturnType = "void"; |
| 572 | return true; |
| 573 | } |
| 574 | |
| 575 | // FIXME: add support for adding other function return types besides void. |
| 576 | // FIXME: assign the value returned by non void extracted function. |
| 577 | llvm::Expected<NewFunction> getExtractedFunction(ExtractionZone &ExtZone, |
| 578 | const SourceManager &SM, |
| 579 | const LangOptions &LangOpts) { |
| 580 | CapturedZoneInfo CapturedInfo = captureZoneInfo(ExtZone); |
| 581 | // Bail out if any break of continue exists |
Shaurya Gupta | 3b08a61f | 2019-08-30 09:57:56 +0000 | [diff] [blame] | 582 | if (CapturedInfo.BrokenControlFlow) |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 583 | return llvm::createStringError(llvm::inconvertibleErrorCode(), |
Shaurya Gupta | 3b08a61f | 2019-08-30 09:57:56 +0000 | [diff] [blame] | 584 | +"Cannot extract break/continue without " |
| 585 | "corresponding loop/switch statement."); |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 586 | NewFunction ExtractedFunc(getSemicolonPolicy(ExtZone, SM, LangOpts)); |
| 587 | ExtractedFunc.BodyRange = ExtZone.ZoneRange; |
| 588 | ExtractedFunc.InsertionPoint = ExtZone.getInsertionPoint(); |
| 589 | ExtractedFunc.EnclosingFuncContext = |
| 590 | ExtZone.EnclosingFunction->getDeclContext(); |
| 591 | if (!createParameters(ExtractedFunc, CapturedInfo) || |
| 592 | !generateReturnProperties(ExtractedFunc, CapturedInfo)) |
| 593 | return llvm::createStringError(llvm::inconvertibleErrorCode(), |
| 594 | +"Too complex to extract."); |
| 595 | return ExtractedFunc; |
| 596 | } |
| 597 | |
| 598 | class ExtractFunction : public Tweak { |
| 599 | public: |
| 600 | const char *id() const override final; |
| 601 | bool prepare(const Selection &Inputs) override; |
| 602 | Expected<Effect> apply(const Selection &Inputs) override; |
| 603 | std::string title() const override { return "Extract to function"; } |
| 604 | Intent intent() const override { return Refactor; } |
| 605 | |
| 606 | private: |
| 607 | ExtractionZone ExtZone; |
| 608 | }; |
| 609 | |
| 610 | REGISTER_TWEAK(ExtractFunction) |
| 611 | tooling::Replacement replaceWithFuncCall(const NewFunction &ExtractedFunc, |
| 612 | const SourceManager &SM, |
| 613 | const LangOptions &LangOpts) { |
| 614 | std::string FuncCall = ExtractedFunc.renderCall(); |
| 615 | return tooling::Replacement( |
| 616 | SM, CharSourceRange(ExtractedFunc.BodyRange, false), FuncCall, LangOpts); |
| 617 | } |
| 618 | |
| 619 | tooling::Replacement createFunctionDefinition(const NewFunction &ExtractedFunc, |
| 620 | const SourceManager &SM) { |
| 621 | std::string FunctionDef = ExtractedFunc.renderDefinition(SM); |
| 622 | return tooling::Replacement(SM, ExtractedFunc.InsertionPoint, 0, FunctionDef); |
| 623 | } |
| 624 | |
| 625 | bool ExtractFunction::prepare(const Selection &Inputs) { |
| 626 | const Node *CommonAnc = Inputs.ASTSelection.commonAncestor(); |
| 627 | const SourceManager &SM = Inputs.AST.getSourceManager(); |
| 628 | const LangOptions &LangOpts = Inputs.AST.getASTContext().getLangOpts(); |
| 629 | if (auto MaybeExtZone = findExtractionZone(CommonAnc, SM, LangOpts)) { |
| 630 | ExtZone = std::move(*MaybeExtZone); |
| 631 | return true; |
| 632 | } |
| 633 | return false; |
| 634 | } |
| 635 | |
| 636 | Expected<Tweak::Effect> ExtractFunction::apply(const Selection &Inputs) { |
| 637 | const SourceManager &SM = Inputs.AST.getSourceManager(); |
| 638 | const LangOptions &LangOpts = Inputs.AST.getASTContext().getLangOpts(); |
| 639 | auto ExtractedFunc = getExtractedFunction(ExtZone, SM, LangOpts); |
| 640 | // FIXME: Add more types of errors. |
| 641 | if (!ExtractedFunc) |
| 642 | return ExtractedFunc.takeError(); |
| 643 | tooling::Replacements Result; |
| 644 | if (auto Err = Result.add(createFunctionDefinition(*ExtractedFunc, SM))) |
| 645 | return std::move(Err); |
| 646 | if (auto Err = Result.add(replaceWithFuncCall(*ExtractedFunc, SM, LangOpts))) |
| 647 | return std::move(Err); |
Kadir Cetinkaya | 5b27093 | 2019-09-09 12:28:44 +0000 | [diff] [blame^] | 648 | return Effect::mainFileEdit(SM, std::move(Result)); |
Shaurya Gupta | bf47734 | 2019-08-28 19:34:17 +0000 | [diff] [blame] | 649 | } |
| 650 | |
| 651 | } // namespace |
| 652 | } // namespace clangd |
| 653 | } // namespace clang |