| 1 | //===--------- IncrementalParser.cpp - Incremental Compilation -----------===// |
| 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 | // This file implements the class which performs incremental code compilation. |
| 10 | // |
| 11 | //===----------------------------------------------------------------------===// |
| 12 | |
| 13 | #include "IncrementalParser.h" |
| 14 | #include "IncrementalAction.h" |
| 15 | |
| 16 | #include "clang/AST/ASTContext.h" |
| 17 | #include "clang/AST/Decl.h" |
| 18 | #include "clang/AST/DeclContextInternals.h" |
| 19 | #include "clang/Frontend/CompilerInstance.h" |
| 20 | #include "clang/Interpreter/PartialTranslationUnit.h" |
| 21 | #include "clang/Parse/Parser.h" |
| 22 | #include "clang/Sema/Sema.h" |
| 23 | #include "llvm/ADT/SmallVector.h" |
| 24 | #include "llvm/IR/Module.h" |
| 25 | #include "llvm/Support/CrashRecoveryContext.h" |
| 26 | #include "llvm/Support/Error.h" |
| 27 | |
| 28 | #include <sstream> |
| 29 | |
| 30 | #define DEBUG_TYPE "clang-repl" |
| 31 | |
| 32 | namespace clang { |
| 33 | |
| 34 | // IncrementalParser::IncrementalParser() {} |
| 35 | |
| 36 | IncrementalParser::IncrementalParser(CompilerInstance &Instance, |
| 37 | IncrementalAction *Act, llvm::Error &Err, |
| 38 | std::list<PartialTranslationUnit> &PTUs) |
| 39 | : S(Instance.getSema()), Act(Act), PTUs(PTUs) { |
| 40 | llvm::ErrorAsOutParameter EAO(&Err); |
| 41 | Consumer = &S.getASTConsumer(); |
| 42 | P.reset(p: new Parser(S.getPreprocessor(), S, /*SkipBodies=*/false)); |
| 43 | |
| 44 | if (ExternalASTSource *External = S.getASTContext().getExternalSource()) |
| 45 | External->StartTranslationUnit(Consumer); |
| 46 | |
| 47 | P->Initialize(); |
| 48 | } |
| 49 | |
| 50 | IncrementalParser::~IncrementalParser() { P.reset(); } |
| 51 | |
| 52 | llvm::Expected<TranslationUnitDecl *> |
| 53 | IncrementalParser::ParseOrWrapTopLevelDecl() { |
| 54 | // Recover resources if we crash before exiting this method. |
| 55 | llvm::CrashRecoveryContextCleanupRegistrar<Sema> CleanupSema(&S); |
| 56 | Sema::GlobalEagerInstantiationScope GlobalInstantiations(S, /*Enabled=*/true, |
| 57 | /*AtEndOfTU=*/true); |
| 58 | Sema::LocalEagerInstantiationScope LocalInstantiations(S, /*AtEndOfTU=*/true); |
| 59 | |
| 60 | // Add a new PTU. |
| 61 | ASTContext &C = S.getASTContext(); |
| 62 | C.addTranslationUnitDecl(); |
| 63 | |
| 64 | // Skip previous eof due to last incremental input. |
| 65 | if (P->getCurToken().is(K: tok::annot_repl_input_end)) { |
| 66 | P->ConsumeAnyToken(); |
| 67 | // FIXME: Clang does not call ExitScope on finalizing the regular TU, we |
| 68 | // might want to do that around HandleEndOfTranslationUnit. |
| 69 | P->ExitScope(); |
| 70 | S.CurContext = nullptr; |
| 71 | // Start a new PTU. |
| 72 | P->EnterScope(ScopeFlags: Scope::DeclScope); |
| 73 | S.ActOnTranslationUnitScope(S: P->getCurScope()); |
| 74 | } |
| 75 | |
| 76 | Parser::DeclGroupPtrTy ADecl; |
| 77 | Sema::ModuleImportState ImportState; |
| 78 | for (bool AtEOF = P->ParseFirstTopLevelDecl(Result&: ADecl, ImportState); !AtEOF; |
| 79 | AtEOF = P->ParseTopLevelDecl(Result&: ADecl, ImportState)) { |
| 80 | if (ADecl && !Consumer->HandleTopLevelDecl(D: ADecl.get())) |
| 81 | return llvm::make_error<llvm::StringError>(Args: "Parsing failed. " |
| 82 | "The consumer rejected a decl" , |
| 83 | Args: std::error_code()); |
| 84 | } |
| 85 | |
| 86 | DiagnosticsEngine &Diags = S.getDiagnostics(); |
| 87 | if (Diags.hasErrorOccurred()) { |
| 88 | CleanUpPTU(MostRecentTU: C.getTranslationUnitDecl()); |
| 89 | |
| 90 | Diags.Reset(/*soft=*/true); |
| 91 | Diags.getClient()->clear(); |
| 92 | return llvm::make_error<llvm::StringError>(Args: "Parsing failed." , |
| 93 | Args: std::error_code()); |
| 94 | } |
| 95 | |
| 96 | // Process any TopLevelDecls generated by #pragma weak. |
| 97 | for (Decl *D : S.WeakTopLevelDecls()) { |
| 98 | DeclGroupRef DGR(D); |
| 99 | Consumer->HandleTopLevelDecl(D: DGR); |
| 100 | } |
| 101 | |
| 102 | LocalInstantiations.perform(); |
| 103 | GlobalInstantiations.perform(); |
| 104 | |
| 105 | Consumer->HandleTranslationUnit(Ctx&: C); |
| 106 | |
| 107 | return C.getTranslationUnitDecl(); |
| 108 | } |
| 109 | |
| 110 | llvm::Expected<TranslationUnitDecl *> |
| 111 | IncrementalParser::Parse(llvm::StringRef input) { |
| 112 | Preprocessor &PP = S.getPreprocessor(); |
| 113 | assert(PP.isIncrementalProcessingEnabled() && "Not in incremental mode!?" ); |
| 114 | |
| 115 | std::ostringstream SourceName; |
| 116 | SourceName << "input_line_" << InputCount++; |
| 117 | |
| 118 | // Create an uninitialized memory buffer, copy code in and append "\n" |
| 119 | size_t InputSize = input.size(); // don't include trailing 0 |
| 120 | // MemBuffer size should *not* include terminating zero |
| 121 | std::unique_ptr<llvm::MemoryBuffer> MB( |
| 122 | llvm::WritableMemoryBuffer::getNewUninitMemBuffer(Size: InputSize + 1, |
| 123 | BufferName: SourceName.str())); |
| 124 | char *MBStart = const_cast<char *>(MB->getBufferStart()); |
| 125 | memcpy(dest: MBStart, src: input.data(), n: InputSize); |
| 126 | MBStart[InputSize] = '\n'; |
| 127 | |
| 128 | SourceManager &SM = S.getSourceManager(); |
| 129 | |
| 130 | // FIXME: Create SourceLocation, which will allow clang to order the overload |
| 131 | // candidates for example |
| 132 | SourceLocation NewLoc = SM.getLocForStartOfFile(FID: SM.getMainFileID()); |
| 133 | |
| 134 | // Create FileID for the current buffer. |
| 135 | FileID FID; |
| 136 | // Create FileEntry and FileID for the current buffer. |
| 137 | FileEntryRef FE = SM.getFileManager().getVirtualFileRef( |
| 138 | Filename: SourceName.str(), Size: InputSize, ModificationTime: 0 /* mod time*/); |
| 139 | SM.overrideFileContents(SourceFile: FE, Buffer: std::move(MB)); |
| 140 | |
| 141 | // Ensure HeaderFileInfo exists before lookup to prevent assertion |
| 142 | HeaderSearch &HS = PP.getHeaderSearchInfo(); |
| 143 | HS.getFileInfo(FE); |
| 144 | |
| 145 | FID = SM.createFileID(SourceFile: FE, IncludePos: NewLoc, FileCharacter: SrcMgr::C_User); |
| 146 | |
| 147 | // NewLoc only used for diags. |
| 148 | if (PP.EnterSourceFile(FID, /*DirLookup=*/Dir: nullptr, Loc: NewLoc)) |
| 149 | return llvm::make_error<llvm::StringError>(Args: "Parsing failed. " |
| 150 | "Cannot enter source file." , |
| 151 | Args: std::error_code()); |
| 152 | |
| 153 | auto PTU = ParseOrWrapTopLevelDecl(); |
| 154 | if (!PTU) |
| 155 | return PTU.takeError(); |
| 156 | |
| 157 | if (PP.getLangOpts().DelayedTemplateParsing) { |
| 158 | // Microsoft-specific: |
| 159 | // Late parsed templates can leave unswallowed "macro"-like tokens. |
| 160 | // They will seriously confuse the Parser when entering the next |
| 161 | // source file. So lex until we are EOF. |
| 162 | Token Tok; |
| 163 | do { |
| 164 | PP.Lex(Result&: Tok); |
| 165 | } while (Tok.isNot(K: tok::annot_repl_input_end)); |
| 166 | } else { |
| 167 | Token AssertTok; |
| 168 | PP.Lex(Result&: AssertTok); |
| 169 | assert(AssertTok.is(tok::annot_repl_input_end) && |
| 170 | "Lexer must be EOF when starting incremental parse!" ); |
| 171 | } |
| 172 | |
| 173 | return PTU; |
| 174 | } |
| 175 | |
| 176 | void IncrementalParser::withdrawMostRecentTU( |
| 177 | TranslationUnitDecl *MostRecentTU) { |
| 178 | TranslationUnitDecl *Prev = MostRecentTU->getPreviousDecl(); |
| 179 | if (!Prev) |
| 180 | return; |
| 181 | assert(MostRecentTU->getMostRecentDecl() == MostRecentTU && |
| 182 | "Not the most recent translation unit!" ); |
| 183 | |
| 184 | // Rebuild A -> ... -> Prev -> MostRecentTU as A -> ... -> Prev. |
| 185 | MostRecentTU->getFirstDecl()->RedeclLink.setLatest(Prev); |
| 186 | |
| 187 | // getTranslationUnitDecl() requires the active unit to be the latest one. |
| 188 | ASTContext &C = S.getASTContext(); |
| 189 | if (C.TraversalScope.size() == 1 && C.TraversalScope.back() == MostRecentTU) |
| 190 | C.TraversalScope = {Prev}; |
| 191 | C.TUDecl = Prev; |
| 192 | } |
| 193 | |
| 194 | void IncrementalParser::CleanUpPTU(TranslationUnitDecl *MostRecentTU) { |
| 195 | if (StoredDeclsMap *Map = MostRecentTU->getPrimaryContext()->getLookupPtr()) { |
| 196 | // Collect the keys to erase: erasing during iteration invalidates the map |
| 197 | // iterator under backward-shift deletion. |
| 198 | llvm::SmallVector<DeclarationName, 16> KeysToErase; |
| 199 | for (auto &&[Key, List] : *Map) { |
| 200 | DeclContextLookupResult R = List.getLookupResult(); |
| 201 | std::vector<NamedDecl *> NamedDeclsToRemove; |
| 202 | bool RemoveAll = true; |
| 203 | for (NamedDecl *D : R) { |
| 204 | if (D->getTranslationUnitDecl() == MostRecentTU) |
| 205 | NamedDeclsToRemove.push_back(x: D); |
| 206 | else |
| 207 | RemoveAll = false; |
| 208 | } |
| 209 | if (LLVM_LIKELY(RemoveAll)) { |
| 210 | KeysToErase.push_back(Elt: Key); |
| 211 | } else { |
| 212 | for (NamedDecl *D : NamedDeclsToRemove) |
| 213 | List.remove(D); |
| 214 | } |
| 215 | } |
| 216 | for (DeclarationName Key : KeysToErase) |
| 217 | Map->erase(Val: Key); |
| 218 | } |
| 219 | |
| 220 | // Check if we need to clean up the IdResolver chain. |
| 221 | auto RemoveFromIdResolver = [&](NamedDecl *D) { |
| 222 | if (D->getDeclName().getFETokenInfo() && !D->getLangOpts().ObjC && |
| 223 | !D->getLangOpts().CPlusPlus) |
| 224 | S.IdResolver.RemoveDecl(D); |
| 225 | }; |
| 226 | |
| 227 | ExternCContextDecl *ECCD = S.getASTContext().getExternCContextDecl(); |
| 228 | if (StoredDeclsMap *Map = ECCD->getPrimaryContext()->getLookupPtr()) { |
| 229 | for (auto &&[Key, List] : *Map) { |
| 230 | DeclContextLookupResult R = List.getLookupResult(); |
| 231 | llvm::SmallVector<NamedDecl *, 4> NamedDeclsToRemove; |
| 232 | for (NamedDecl *D : R) { |
| 233 | // Implicitly generated C decl is not attached to the current TU but |
| 234 | // lexically attached to the recent TU, so we need to check the lexical |
| 235 | // context. |
| 236 | DeclContext *LDC = D->getLexicalDeclContext(); |
| 237 | while (LDC && !isa<TranslationUnitDecl>(Val: LDC)) |
| 238 | LDC = LDC->getLexicalParent(); |
| 239 | TranslationUnitDecl *TopTU = cast_or_null<TranslationUnitDecl>(Val: LDC); |
| 240 | if (TopTU == MostRecentTU) |
| 241 | NamedDeclsToRemove.push_back(Elt: D); |
| 242 | } |
| 243 | for (NamedDecl *D : NamedDeclsToRemove) { |
| 244 | List.remove(D); |
| 245 | RemoveFromIdResolver(D); |
| 246 | } |
| 247 | } |
| 248 | } |
| 249 | |
| 250 | for (Decl *D : MostRecentTU->decls()) { |
| 251 | auto *ND = dyn_cast<NamedDecl>(Val: D); |
| 252 | if (!ND || ND->getDeclName().isEmpty()) |
| 253 | continue; |
| 254 | RemoveFromIdResolver(ND); |
| 255 | } |
| 256 | |
| 257 | // Lookup alone is not enough: the redeclaration chain still reaches these. |
| 258 | withdrawMostRecentTU(MostRecentTU); |
| 259 | } |
| 260 | |
| 261 | PartialTranslationUnit & |
| 262 | IncrementalParser::RegisterPTU(TranslationUnitDecl *TU, |
| 263 | std::unique_ptr<llvm::Module> M /*={}*/) { |
| 264 | PTUs.emplace_back(args: PartialTranslationUnit()); |
| 265 | PartialTranslationUnit &LastPTU = PTUs.back(); |
| 266 | LastPTU.TUPart = TU; |
| 267 | |
| 268 | if (!M) |
| 269 | M = Act->GenModule(); |
| 270 | |
| 271 | assert((!Act->getCodeGen() || M) && "Must have a llvm::Module at this point" ); |
| 272 | |
| 273 | LastPTU.TheModule = std::move(M); |
| 274 | LLVM_DEBUG(llvm::dbgs() << "compile-ptu " << PTUs.size() - 1 |
| 275 | << ": [TU=" << LastPTU.TUPart); |
| 276 | if (LastPTU.TheModule) |
| 277 | LLVM_DEBUG(llvm::dbgs() << ", M=" << LastPTU.TheModule.get() << " (" |
| 278 | << LastPTU.TheModule->getName() << ")" ); |
| 279 | LLVM_DEBUG(llvm::dbgs() << "]\n" ); |
| 280 | return LastPTU; |
| 281 | } |
| 282 | } // end namespace clang |
| 283 | |