1 //===--- HeaderIncludes.cpp - Insert/Delete #includes --*- 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 #include "clang/Tooling/Inclusions/HeaderIncludes.h" 10 #include "clang/Basic/FileManager.h" 11 #include "clang/Basic/SourceManager.h" 12 #include "clang/Lex/Lexer.h" 13 #include "llvm/Support/FormatVariadic.h" 14 #include "llvm/Support/Path.h" 15 #include <optional> 16 17 namespace clang { 18 namespace tooling { 19 namespace { 20 21 LangOptions createLangOpts() { 22 LangOptions LangOpts; 23 LangOpts.CPlusPlus = 1; 24 LangOpts.CPlusPlus11 = 1; 25 LangOpts.CPlusPlus14 = 1; 26 LangOpts.LineComment = 1; 27 LangOpts.CXXOperatorNames = 1; 28 LangOpts.Bool = 1; 29 LangOpts.ObjC = 1; 30 LangOpts.MicrosoftExt = 1; // To get kw___try, kw___finally. 31 LangOpts.DeclSpecKeyword = 1; // To get __declspec. 32 LangOpts.WChar = 1; // To get wchar_t 33 return LangOpts; 34 } 35 36 // Returns the offset after skipping a sequence of tokens, matched by \p 37 // GetOffsetAfterSequence, from the start of the code. 38 // \p GetOffsetAfterSequence should be a function that matches a sequence of 39 // tokens and returns an offset after the sequence. 40 unsigned getOffsetAfterTokenSequence( 41 StringRef FileName, StringRef Code, const IncludeStyle &Style, 42 llvm::function_ref<unsigned(const SourceManager &, Lexer &, Token &)> 43 GetOffsetAfterSequence) { 44 SourceManagerForFile VirtualSM(FileName, Code); 45 SourceManager &SM = VirtualSM.get(); 46 LangOptions LangOpts = createLangOpts(); 47 Lexer Lex(SM.getMainFileID(), SM.getBufferOrFake(SM.getMainFileID()), SM, 48 LangOpts); 49 Token Tok; 50 // Get the first token. 51 Lex.LexFromRawLexer(Tok); 52 return GetOffsetAfterSequence(SM, Lex, Tok); 53 } 54 55 // Check if a sequence of tokens is like "#<Name> <raw_identifier>". If it is, 56 // \p Tok will be the token after this directive; otherwise, it can be any token 57 // after the given \p Tok (including \p Tok). If \p RawIDName is provided, the 58 // (second) raw_identifier name is checked. 59 bool checkAndConsumeDirectiveWithName( 60 Lexer &Lex, StringRef Name, Token &Tok, 61 std::optional<StringRef> RawIDName = std::nullopt) { 62 bool Matched = Tok.is(tok::hash) && !Lex.LexFromRawLexer(Tok) && 63 Tok.is(tok::raw_identifier) && 64 Tok.getRawIdentifier() == Name && !Lex.LexFromRawLexer(Tok) && 65 Tok.is(tok::raw_identifier) && 66 (!RawIDName || Tok.getRawIdentifier() == *RawIDName); 67 if (Matched) 68 Lex.LexFromRawLexer(Tok); 69 return Matched; 70 } 71 72 void skipComments(Lexer &Lex, Token &Tok) { 73 while (Tok.is(tok::comment)) 74 if (Lex.LexFromRawLexer(Tok)) 75 return; 76 } 77 78 // Returns the offset after header guard directives and any comments 79 // before/after header guards (e.g. #ifndef/#define pair, #pragma once). If no 80 // header guard is present in the code, this will return the offset after 81 // skipping all comments from the start of the code. 82 unsigned getOffsetAfterHeaderGuardsAndComments(StringRef FileName, 83 StringRef Code, 84 const IncludeStyle &Style) { 85 // \p Consume returns location after header guard or 0 if no header guard is 86 // found. 87 auto ConsumeHeaderGuardAndComment = 88 [&](std::function<unsigned(const SourceManager &SM, Lexer &Lex, 89 Token Tok)> 90 Consume) { 91 return getOffsetAfterTokenSequence( 92 FileName, Code, Style, 93 [&Consume](const SourceManager &SM, Lexer &Lex, Token Tok) { 94 skipComments(Lex, Tok); 95 unsigned InitialOffset = SM.getFileOffset(Tok.getLocation()); 96 return std::max(InitialOffset, Consume(SM, Lex, Tok)); 97 }); 98 }; 99 return std::max( 100 // #ifndef/#define 101 ConsumeHeaderGuardAndComment( 102 [](const SourceManager &SM, Lexer &Lex, Token Tok) -> unsigned { 103 if (checkAndConsumeDirectiveWithName(Lex, "ifndef", Tok)) { 104 skipComments(Lex, Tok); 105 if (checkAndConsumeDirectiveWithName(Lex, "define", Tok) && 106 Tok.isAtStartOfLine()) 107 return SM.getFileOffset(Tok.getLocation()); 108 } 109 return 0; 110 }), 111 // #pragma once 112 ConsumeHeaderGuardAndComment( 113 [](const SourceManager &SM, Lexer &Lex, Token Tok) -> unsigned { 114 if (checkAndConsumeDirectiveWithName(Lex, "pragma", Tok, 115 StringRef("once"))) 116 return SM.getFileOffset(Tok.getLocation()); 117 return 0; 118 })); 119 } 120 121 // Check if a sequence of tokens is like 122 // "#include ("header.h" | <header.h>)". 123 // If it is, \p Tok will be the token after this directive; otherwise, it can be 124 // any token after the given \p Tok (including \p Tok). 125 bool checkAndConsumeInclusiveDirective(Lexer &Lex, Token &Tok) { 126 auto Matched = [&]() { 127 Lex.LexFromRawLexer(Tok); 128 return true; 129 }; 130 if (Tok.is(tok::hash) && !Lex.LexFromRawLexer(Tok) && 131 Tok.is(tok::raw_identifier) && Tok.getRawIdentifier() == "include") { 132 if (Lex.LexFromRawLexer(Tok)) 133 return false; 134 if (Tok.is(tok::string_literal)) 135 return Matched(); 136 if (Tok.is(tok::less)) { 137 while (!Lex.LexFromRawLexer(Tok) && Tok.isNot(tok::greater)) { 138 } 139 if (Tok.is(tok::greater)) 140 return Matched(); 141 } 142 } 143 return false; 144 } 145 146 // Returns the offset of the last #include directive after which a new 147 // #include can be inserted. This ignores #include's after the #include block(s) 148 // in the beginning of a file to avoid inserting headers into code sections 149 // where new #include's should not be added by default. 150 // These code sections include: 151 // - raw string literals (containing #include). 152 // - #if blocks. 153 // - Special #include's among declarations (e.g. functions). 154 // 155 // If no #include after which a new #include can be inserted, this returns the 156 // offset after skipping all comments from the start of the code. 157 // Inserting after an #include is not allowed if it comes after code that is not 158 // #include (e.g. pre-processing directive that is not #include, declarations). 159 unsigned getMaxHeaderInsertionOffset(StringRef FileName, StringRef Code, 160 const IncludeStyle &Style) { 161 return getOffsetAfterTokenSequence( 162 FileName, Code, Style, 163 [](const SourceManager &SM, Lexer &Lex, Token Tok) { 164 skipComments(Lex, Tok); 165 unsigned MaxOffset = SM.getFileOffset(Tok.getLocation()); 166 while (checkAndConsumeInclusiveDirective(Lex, Tok)) 167 MaxOffset = SM.getFileOffset(Tok.getLocation()); 168 return MaxOffset; 169 }); 170 } 171 172 inline StringRef trimInclude(StringRef IncludeName) { 173 return IncludeName.trim("\"<>"); 174 } 175 176 const char IncludeRegexPattern[] = 177 R"(^[\t\ ]*#[\t\ ]*(import|include)[^"<]*(["<][^">]*[">]))"; 178 179 // The filename of Path excluding extension. 180 // Used to match implementation with headers, this differs from sys::path::stem: 181 // - in names with multiple dots (foo.cu.cc) it terminates at the *first* 182 // - an empty stem is never returned: /foo/.bar.x => .bar 183 // - we don't bother to handle . and .. specially 184 StringRef matchingStem(llvm::StringRef Path) { 185 StringRef Name = llvm::sys::path::filename(Path); 186 return Name.substr(0, Name.find('.', 1)); 187 } 188 189 } // anonymous namespace 190 191 IncludeCategoryManager::IncludeCategoryManager(const IncludeStyle &Style, 192 StringRef FileName) 193 : Style(Style), FileName(FileName) { 194 for (const auto &Category : Style.IncludeCategories) { 195 CategoryRegexs.emplace_back(Category.Regex, Category.RegexIsCaseSensitive 196 ? llvm::Regex::NoFlags 197 : llvm::Regex::IgnoreCase); 198 } 199 IsMainFile = FileName.endswith(".c") || FileName.endswith(".cc") || 200 FileName.endswith(".cpp") || FileName.endswith(".c++") || 201 FileName.endswith(".cxx") || FileName.endswith(".m") || 202 FileName.endswith(".mm"); 203 if (!Style.IncludeIsMainSourceRegex.empty()) { 204 llvm::Regex MainFileRegex(Style.IncludeIsMainSourceRegex); 205 IsMainFile |= MainFileRegex.match(FileName); 206 } 207 } 208 209 int IncludeCategoryManager::getIncludePriority(StringRef IncludeName, 210 bool CheckMainHeader) const { 211 int Ret = INT_MAX; 212 for (unsigned i = 0, e = CategoryRegexs.size(); i != e; ++i) 213 if (CategoryRegexs[i].match(IncludeName)) { 214 Ret = Style.IncludeCategories[i].Priority; 215 break; 216 } 217 if (CheckMainHeader && IsMainFile && Ret > 0 && isMainHeader(IncludeName)) 218 Ret = 0; 219 return Ret; 220 } 221 222 int IncludeCategoryManager::getSortIncludePriority(StringRef IncludeName, 223 bool CheckMainHeader) const { 224 int Ret = INT_MAX; 225 for (unsigned i = 0, e = CategoryRegexs.size(); i != e; ++i) 226 if (CategoryRegexs[i].match(IncludeName)) { 227 Ret = Style.IncludeCategories[i].SortPriority; 228 if (Ret == 0) 229 Ret = Style.IncludeCategories[i].Priority; 230 break; 231 } 232 if (CheckMainHeader && IsMainFile && Ret > 0 && isMainHeader(IncludeName)) 233 Ret = 0; 234 return Ret; 235 } 236 bool IncludeCategoryManager::isMainHeader(StringRef IncludeName) const { 237 if (!IncludeName.startswith("\"")) 238 return false; 239 240 IncludeName = 241 IncludeName.drop_front(1).drop_back(1); // remove the surrounding "" or <> 242 // Not matchingStem: implementation files may have compound extensions but 243 // headers may not. 244 StringRef HeaderStem = llvm::sys::path::stem(IncludeName); 245 StringRef FileStem = llvm::sys::path::stem(FileName); // foo.cu for foo.cu.cc 246 StringRef MatchingFileStem = matchingStem(FileName); // foo for foo.cu.cc 247 // main-header examples: 248 // 1) foo.h => foo.cc 249 // 2) foo.h => foo.cu.cc 250 // 3) foo.proto.h => foo.proto.cc 251 // 252 // non-main-header examples: 253 // 1) foo.h => bar.cc 254 // 2) foo.proto.h => foo.cc 255 StringRef Matching; 256 if (MatchingFileStem.starts_with_insensitive(HeaderStem)) 257 Matching = MatchingFileStem; // example 1), 2) 258 else if (FileStem.equals_insensitive(HeaderStem)) 259 Matching = FileStem; // example 3) 260 if (!Matching.empty()) { 261 llvm::Regex MainIncludeRegex(HeaderStem.str() + Style.IncludeIsMainRegex, 262 llvm::Regex::IgnoreCase); 263 if (MainIncludeRegex.match(Matching)) 264 return true; 265 } 266 return false; 267 } 268 269 const llvm::Regex HeaderIncludes::IncludeRegex(IncludeRegexPattern); 270 271 HeaderIncludes::HeaderIncludes(StringRef FileName, StringRef Code, 272 const IncludeStyle &Style) 273 : FileName(FileName), Code(Code), FirstIncludeOffset(-1), 274 MinInsertOffset( 275 getOffsetAfterHeaderGuardsAndComments(FileName, Code, Style)), 276 MaxInsertOffset(MinInsertOffset + 277 getMaxHeaderInsertionOffset( 278 FileName, Code.drop_front(MinInsertOffset), Style)), 279 MainIncludeFound(false), 280 Categories(Style, FileName) { 281 // Add 0 for main header and INT_MAX for headers that are not in any 282 // category. 283 Priorities = {0, INT_MAX}; 284 for (const auto &Category : Style.IncludeCategories) 285 Priorities.insert(Category.Priority); 286 SmallVector<StringRef, 32> Lines; 287 Code.drop_front(MinInsertOffset).split(Lines, "\n"); 288 289 unsigned Offset = MinInsertOffset; 290 unsigned NextLineOffset; 291 SmallVector<StringRef, 4> Matches; 292 for (auto Line : Lines) { 293 NextLineOffset = std::min(Code.size(), Offset + Line.size() + 1); 294 if (IncludeRegex.match(Line, &Matches)) { 295 // If this is the last line without trailing newline, we need to make 296 // sure we don't delete across the file boundary. 297 addExistingInclude( 298 Include(Matches[2], 299 tooling::Range( 300 Offset, std::min(Line.size() + 1, Code.size() - Offset)), 301 Matches[1] == "import" ? tooling::IncludeDirective::Import 302 : tooling::IncludeDirective::Include), 303 NextLineOffset); 304 } 305 Offset = NextLineOffset; 306 } 307 308 // Populate CategoryEndOfssets: 309 // - Ensure that CategoryEndOffset[Highest] is always populated. 310 // - If CategoryEndOffset[Priority] isn't set, use the next higher value 311 // that is set, up to CategoryEndOffset[Highest]. 312 auto Highest = Priorities.begin(); 313 if (CategoryEndOffsets.find(*Highest) == CategoryEndOffsets.end()) { 314 if (FirstIncludeOffset >= 0) 315 CategoryEndOffsets[*Highest] = FirstIncludeOffset; 316 else 317 CategoryEndOffsets[*Highest] = MinInsertOffset; 318 } 319 // By this point, CategoryEndOffset[Highest] is always set appropriately: 320 // - to an appropriate location before/after existing #includes, or 321 // - to right after the header guard, or 322 // - to the beginning of the file. 323 for (auto I = ++Priorities.begin(), E = Priorities.end(); I != E; ++I) 324 if (CategoryEndOffsets.find(*I) == CategoryEndOffsets.end()) 325 CategoryEndOffsets[*I] = CategoryEndOffsets[*std::prev(I)]; 326 } 327 328 // \p Offset: the start of the line following this include directive. 329 void HeaderIncludes::addExistingInclude(Include IncludeToAdd, 330 unsigned NextLineOffset) { 331 auto Iter = 332 ExistingIncludes.try_emplace(trimInclude(IncludeToAdd.Name)).first; 333 Iter->second.push_back(std::move(IncludeToAdd)); 334 auto &CurInclude = Iter->second.back(); 335 // The header name with quotes or angle brackets. 336 // Only record the offset of current #include if we can insert after it. 337 if (CurInclude.R.getOffset() <= MaxInsertOffset) { 338 int Priority = Categories.getIncludePriority( 339 CurInclude.Name, /*CheckMainHeader=*/!MainIncludeFound); 340 if (Priority == 0) 341 MainIncludeFound = true; 342 CategoryEndOffsets[Priority] = NextLineOffset; 343 IncludesByPriority[Priority].push_back(&CurInclude); 344 if (FirstIncludeOffset < 0) 345 FirstIncludeOffset = CurInclude.R.getOffset(); 346 } 347 } 348 349 std::optional<tooling::Replacement> 350 HeaderIncludes::insert(llvm::StringRef IncludeName, bool IsAngled, 351 IncludeDirective Directive) const { 352 assert(IncludeName == trimInclude(IncludeName)); 353 // If a <header> ("header") already exists in code, "header" (<header>) with 354 // different quotation and/or directive will still be inserted. 355 // FIXME: figure out if this is the best behavior. 356 auto It = ExistingIncludes.find(IncludeName); 357 if (It != ExistingIncludes.end()) { 358 for (const auto &Inc : It->second) 359 if (Inc.Directive == Directive && 360 ((IsAngled && StringRef(Inc.Name).startswith("<")) || 361 (!IsAngled && StringRef(Inc.Name).startswith("\"")))) 362 return std::nullopt; 363 } 364 std::string Quoted = 365 std::string(llvm::formatv(IsAngled ? "<{0}>" : "\"{0}\"", IncludeName)); 366 StringRef QuotedName = Quoted; 367 int Priority = Categories.getIncludePriority( 368 QuotedName, /*CheckMainHeader=*/!MainIncludeFound); 369 auto CatOffset = CategoryEndOffsets.find(Priority); 370 assert(CatOffset != CategoryEndOffsets.end()); 371 unsigned InsertOffset = CatOffset->second; // Fall back offset 372 auto Iter = IncludesByPriority.find(Priority); 373 if (Iter != IncludesByPriority.end()) { 374 for (const auto *Inc : Iter->second) { 375 if (QuotedName < Inc->Name) { 376 InsertOffset = Inc->R.getOffset(); 377 break; 378 } 379 } 380 } 381 assert(InsertOffset <= Code.size()); 382 llvm::StringRef DirectiveSpelling = 383 Directive == IncludeDirective::Include ? "include" : "import"; 384 std::string NewInclude = 385 llvm::formatv("#{0} {1}\n", DirectiveSpelling, QuotedName); 386 // When inserting headers at end of the code, also append '\n' to the code 387 // if it does not end with '\n'. 388 // FIXME: when inserting multiple #includes at the end of code, only one 389 // newline should be added. 390 if (InsertOffset == Code.size() && (!Code.empty() && Code.back() != '\n')) 391 NewInclude = "\n" + NewInclude; 392 return tooling::Replacement(FileName, InsertOffset, 0, NewInclude); 393 } 394 395 tooling::Replacements HeaderIncludes::remove(llvm::StringRef IncludeName, 396 bool IsAngled) const { 397 assert(IncludeName == trimInclude(IncludeName)); 398 tooling::Replacements Result; 399 auto Iter = ExistingIncludes.find(IncludeName); 400 if (Iter == ExistingIncludes.end()) 401 return Result; 402 for (const auto &Inc : Iter->second) { 403 if ((IsAngled && StringRef(Inc.Name).startswith("\"")) || 404 (!IsAngled && StringRef(Inc.Name).startswith("<"))) 405 continue; 406 llvm::Error Err = Result.add(tooling::Replacement( 407 FileName, Inc.R.getOffset(), Inc.R.getLength(), "")); 408 if (Err) { 409 auto ErrMsg = "Unexpected conflicts in #include deletions: " + 410 llvm::toString(std::move(Err)); 411 llvm_unreachable(ErrMsg.c_str()); 412 } 413 } 414 return Result; 415 } 416 417 } // namespace tooling 418 } // namespace clang 419