1 //===--- HeaderIncludes.cpp - Insert/Delete #includes --*- C++ -*----------===//
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
7 //===----------------------------------------------------------------------===//
9 #include "clang/Tooling/Inclusions/HeaderIncludes.h"
10 #include "clang/Basic/SourceManager.h"
11 #include "clang/Lex/Lexer.h"
12 #include "llvm/ADT/Optional.h"
13 #include "llvm/Support/FormatVariadic.h"
19 LangOptions createLangOpts() {
21 LangOpts.CPlusPlus = 1;
22 LangOpts.CPlusPlus11 = 1;
23 LangOpts.CPlusPlus14 = 1;
24 LangOpts.LineComment = 1;
25 LangOpts.CXXOperatorNames = 1;
28 LangOpts.MicrosoftExt = 1; // To get kw___try, kw___finally.
29 LangOpts.DeclSpecKeyword = 1; // To get __declspec.
30 LangOpts.WChar = 1; // To get wchar_t
34 // Returns the offset after skipping a sequence of tokens, matched by \p
35 // GetOffsetAfterSequence, from the start of the code.
36 // \p GetOffsetAfterSequence should be a function that matches a sequence of
37 // tokens and returns an offset after the sequence.
38 unsigned getOffsetAfterTokenSequence(
39 StringRef FileName, StringRef Code, const IncludeStyle &Style,
40 llvm::function_ref<unsigned(const SourceManager &, Lexer &, Token &)>
41 GetOffsetAfterSequence) {
42 SourceManagerForFile VirtualSM(FileName, Code);
43 SourceManager &SM = VirtualSM.get();
44 Lexer Lex(SM.getMainFileID(), SM.getBuffer(SM.getMainFileID()), SM,
47 // Get the first token.
48 Lex.LexFromRawLexer(Tok);
49 return GetOffsetAfterSequence(SM, Lex, Tok);
52 // Check if a sequence of tokens is like "#<Name> <raw_identifier>". If it is,
53 // \p Tok will be the token after this directive; otherwise, it can be any token
54 // after the given \p Tok (including \p Tok). If \p RawIDName is provided, the
55 // (second) raw_identifier name is checked.
56 bool checkAndConsumeDirectiveWithName(
57 Lexer &Lex, StringRef Name, Token &Tok,
58 llvm::Optional<StringRef> RawIDName = llvm::None) {
59 bool Matched = Tok.is(tok::hash) && !Lex.LexFromRawLexer(Tok) &&
60 Tok.is(tok::raw_identifier) &&
61 Tok.getRawIdentifier() == Name && !Lex.LexFromRawLexer(Tok) &&
62 Tok.is(tok::raw_identifier) &&
63 (!RawIDName || Tok.getRawIdentifier() == *RawIDName);
65 Lex.LexFromRawLexer(Tok);
69 void skipComments(Lexer &Lex, Token &Tok) {
70 while (Tok.is(tok::comment))
71 if (Lex.LexFromRawLexer(Tok))
75 // Returns the offset after header guard directives and any comments
76 // before/after header guards (e.g. #ifndef/#define pair, #pragma once). If no
77 // header guard is present in the code, this will return the offset after
78 // skipping all comments from the start of the code.
79 unsigned getOffsetAfterHeaderGuardsAndComments(StringRef FileName,
81 const IncludeStyle &Style) {
82 // \p Consume returns location after header guard or 0 if no header guard is
84 auto ConsumeHeaderGuardAndComment =
85 [&](std::function<unsigned(const SourceManager &SM, Lexer &Lex,
88 return getOffsetAfterTokenSequence(
89 FileName, Code, Style,
90 [&Consume](const SourceManager &SM, Lexer &Lex, Token Tok) {
91 skipComments(Lex, Tok);
92 unsigned InitialOffset = SM.getFileOffset(Tok.getLocation());
93 return std::max(InitialOffset, Consume(SM, Lex, Tok));
98 ConsumeHeaderGuardAndComment(
99 [](const SourceManager &SM, Lexer &Lex, Token Tok) -> unsigned {
100 if (checkAndConsumeDirectiveWithName(Lex, "ifndef", Tok)) {
101 skipComments(Lex, Tok);
102 if (checkAndConsumeDirectiveWithName(Lex, "define", Tok))
103 return SM.getFileOffset(Tok.getLocation());
108 ConsumeHeaderGuardAndComment(
109 [](const SourceManager &SM, Lexer &Lex, Token Tok) -> unsigned {
110 if (checkAndConsumeDirectiveWithName(Lex, "pragma", Tok,
112 return SM.getFileOffset(Tok.getLocation());
117 // Check if a sequence of tokens is like
118 // "#include ("header.h" | <header.h>)".
119 // If it is, \p Tok will be the token after this directive; otherwise, it can be
120 // any token after the given \p Tok (including \p Tok).
121 bool checkAndConsumeInclusiveDirective(Lexer &Lex, Token &Tok) {
122 auto Matched = [&]() {
123 Lex.LexFromRawLexer(Tok);
126 if (Tok.is(tok::hash) && !Lex.LexFromRawLexer(Tok) &&
127 Tok.is(tok::raw_identifier) && Tok.getRawIdentifier() == "include") {
128 if (Lex.LexFromRawLexer(Tok))
130 if (Tok.is(tok::string_literal))
132 if (Tok.is(tok::less)) {
133 while (!Lex.LexFromRawLexer(Tok) && Tok.isNot(tok::greater)) {
135 if (Tok.is(tok::greater))
142 // Returns the offset of the last #include directive after which a new
143 // #include can be inserted. This ignores #include's after the #include block(s)
144 // in the beginning of a file to avoid inserting headers into code sections
145 // where new #include's should not be added by default.
146 // These code sections include:
147 // - raw string literals (containing #include).
149 // - Special #include's among declarations (e.g. functions).
151 // If no #include after which a new #include can be inserted, this returns the
152 // offset after skipping all comments from the start of the code.
153 // Inserting after an #include is not allowed if it comes after code that is not
154 // #include (e.g. pre-processing directive that is not #include, declarations).
155 unsigned getMaxHeaderInsertionOffset(StringRef FileName, StringRef Code,
156 const IncludeStyle &Style) {
157 return getOffsetAfterTokenSequence(
158 FileName, Code, Style,
159 [](const SourceManager &SM, Lexer &Lex, Token Tok) {
160 skipComments(Lex, Tok);
161 unsigned MaxOffset = SM.getFileOffset(Tok.getLocation());
162 while (checkAndConsumeInclusiveDirective(Lex, Tok))
163 MaxOffset = SM.getFileOffset(Tok.getLocation());
168 inline StringRef trimInclude(StringRef IncludeName) {
169 return IncludeName.trim("\"<>");
172 const char IncludeRegexPattern[] =
173 R"(^[\t\ ]*#[\t\ ]*(import|include)[^"<]*(["<][^">]*[">]))";
175 } // anonymous namespace
177 IncludeCategoryManager::IncludeCategoryManager(const IncludeStyle &Style,
179 : Style(Style), FileName(FileName) {
180 FileStem = llvm::sys::path::stem(FileName);
181 for (const auto &Category : Style.IncludeCategories)
182 CategoryRegexs.emplace_back(Category.Regex, llvm::Regex::IgnoreCase);
183 IsMainFile = FileName.endswith(".c") || FileName.endswith(".cc") ||
184 FileName.endswith(".cpp") || FileName.endswith(".c++") ||
185 FileName.endswith(".cxx") || FileName.endswith(".m") ||
186 FileName.endswith(".mm");
187 if (!Style.IncludeIsMainSourceRegex.empty()) {
188 llvm::Regex MainFileRegex(Style.IncludeIsMainSourceRegex);
189 IsMainFile |= MainFileRegex.match(FileName);
193 int IncludeCategoryManager::getIncludePriority(StringRef IncludeName,
194 bool CheckMainHeader) const {
196 for (unsigned i = 0, e = CategoryRegexs.size(); i != e; ++i)
197 if (CategoryRegexs[i].match(IncludeName)) {
198 Ret = Style.IncludeCategories[i].Priority;
201 if (CheckMainHeader && IsMainFile && Ret > 0 && isMainHeader(IncludeName))
206 int IncludeCategoryManager::getSortIncludePriority(StringRef IncludeName,
207 bool CheckMainHeader) const {
209 for (unsigned i = 0, e = CategoryRegexs.size(); i != e; ++i)
210 if (CategoryRegexs[i].match(IncludeName)) {
211 Ret = Style.IncludeCategories[i].SortPriority;
213 Ret = Style.IncludeCategories[i].Priority;
216 if (CheckMainHeader && IsMainFile && Ret > 0 && isMainHeader(IncludeName))
220 bool IncludeCategoryManager::isMainHeader(StringRef IncludeName) const {
221 if (!IncludeName.startswith("\""))
223 StringRef HeaderStem =
224 llvm::sys::path::stem(IncludeName.drop_front(1).drop_back(1));
225 if (FileStem.startswith(HeaderStem) ||
226 FileStem.startswith_lower(HeaderStem)) {
227 llvm::Regex MainIncludeRegex(HeaderStem.str() + Style.IncludeIsMainRegex,
228 llvm::Regex::IgnoreCase);
229 if (MainIncludeRegex.match(FileStem))
235 HeaderIncludes::HeaderIncludes(StringRef FileName, StringRef Code,
236 const IncludeStyle &Style)
237 : FileName(FileName), Code(Code), FirstIncludeOffset(-1),
239 getOffsetAfterHeaderGuardsAndComments(FileName, Code, Style)),
240 MaxInsertOffset(MinInsertOffset +
241 getMaxHeaderInsertionOffset(
242 FileName, Code.drop_front(MinInsertOffset), Style)),
243 Categories(Style, FileName),
244 IncludeRegex(llvm::Regex(IncludeRegexPattern)) {
245 // Add 0 for main header and INT_MAX for headers that are not in any
247 Priorities = {0, INT_MAX};
248 for (const auto &Category : Style.IncludeCategories)
249 Priorities.insert(Category.Priority);
250 SmallVector<StringRef, 32> Lines;
251 Code.drop_front(MinInsertOffset).split(Lines, "\n");
253 unsigned Offset = MinInsertOffset;
254 unsigned NextLineOffset;
255 SmallVector<StringRef, 4> Matches;
256 for (auto Line : Lines) {
257 NextLineOffset = std::min(Code.size(), Offset + Line.size() + 1);
258 if (IncludeRegex.match(Line, &Matches)) {
259 // If this is the last line without trailing newline, we need to make
260 // sure we don't delete across the file boundary.
264 Offset, std::min(Line.size() + 1, Code.size() - Offset))),
267 Offset = NextLineOffset;
270 // Populate CategoryEndOfssets:
271 // - Ensure that CategoryEndOffset[Highest] is always populated.
272 // - If CategoryEndOffset[Priority] isn't set, use the next higher value
273 // that is set, up to CategoryEndOffset[Highest].
274 auto Highest = Priorities.begin();
275 if (CategoryEndOffsets.find(*Highest) == CategoryEndOffsets.end()) {
276 if (FirstIncludeOffset >= 0)
277 CategoryEndOffsets[*Highest] = FirstIncludeOffset;
279 CategoryEndOffsets[*Highest] = MinInsertOffset;
281 // By this point, CategoryEndOffset[Highest] is always set appropriately:
282 // - to an appropriate location before/after existing #includes, or
283 // - to right after the header guard, or
284 // - to the beginning of the file.
285 for (auto I = ++Priorities.begin(), E = Priorities.end(); I != E; ++I)
286 if (CategoryEndOffsets.find(*I) == CategoryEndOffsets.end())
287 CategoryEndOffsets[*I] = CategoryEndOffsets[*std::prev(I)];
290 // \p Offset: the start of the line following this include directive.
291 void HeaderIncludes::addExistingInclude(Include IncludeToAdd,
292 unsigned NextLineOffset) {
294 ExistingIncludes.try_emplace(trimInclude(IncludeToAdd.Name)).first;
295 Iter->second.push_back(std::move(IncludeToAdd));
296 auto &CurInclude = Iter->second.back();
297 // The header name with quotes or angle brackets.
298 // Only record the offset of current #include if we can insert after it.
299 if (CurInclude.R.getOffset() <= MaxInsertOffset) {
300 int Priority = Categories.getIncludePriority(
301 CurInclude.Name, /*CheckMainHeader=*/FirstIncludeOffset < 0);
302 CategoryEndOffsets[Priority] = NextLineOffset;
303 IncludesByPriority[Priority].push_back(&CurInclude);
304 if (FirstIncludeOffset < 0)
305 FirstIncludeOffset = CurInclude.R.getOffset();
309 llvm::Optional<tooling::Replacement>
310 HeaderIncludes::insert(llvm::StringRef IncludeName, bool IsAngled) const {
311 assert(IncludeName == trimInclude(IncludeName));
312 // If a <header> ("header") already exists in code, "header" (<header>) with
313 // different quotation will still be inserted.
314 // FIXME: figure out if this is the best behavior.
315 auto It = ExistingIncludes.find(IncludeName);
316 if (It != ExistingIncludes.end())
317 for (const auto &Inc : It->second)
318 if ((IsAngled && StringRef(Inc.Name).startswith("<")) ||
319 (!IsAngled && StringRef(Inc.Name).startswith("\"")))
322 llvm::formatv(IsAngled ? "<{0}>" : "\"{0}\"", IncludeName);
323 StringRef QuotedName = Quoted;
324 int Priority = Categories.getIncludePriority(
325 QuotedName, /*CheckMainHeader=*/FirstIncludeOffset < 0);
326 auto CatOffset = CategoryEndOffsets.find(Priority);
327 assert(CatOffset != CategoryEndOffsets.end());
328 unsigned InsertOffset = CatOffset->second; // Fall back offset
329 auto Iter = IncludesByPriority.find(Priority);
330 if (Iter != IncludesByPriority.end()) {
331 for (const auto *Inc : Iter->second) {
332 if (QuotedName < Inc->Name) {
333 InsertOffset = Inc->R.getOffset();
338 assert(InsertOffset <= Code.size());
339 std::string NewInclude = llvm::formatv("#include {0}\n", QuotedName);
340 // When inserting headers at end of the code, also append '\n' to the code
341 // if it does not end with '\n'.
342 // FIXME: when inserting multiple #includes at the end of code, only one
343 // newline should be added.
344 if (InsertOffset == Code.size() && (!Code.empty() && Code.back() != '\n'))
345 NewInclude = "\n" + NewInclude;
346 return tooling::Replacement(FileName, InsertOffset, 0, NewInclude);
349 tooling::Replacements HeaderIncludes::remove(llvm::StringRef IncludeName,
350 bool IsAngled) const {
351 assert(IncludeName == trimInclude(IncludeName));
352 tooling::Replacements Result;
353 auto Iter = ExistingIncludes.find(IncludeName);
354 if (Iter == ExistingIncludes.end())
356 for (const auto &Inc : Iter->second) {
357 if ((IsAngled && StringRef(Inc.Name).startswith("\"")) ||
358 (!IsAngled && StringRef(Inc.Name).startswith("<")))
360 llvm::Error Err = Result.add(tooling::Replacement(
361 FileName, Inc.R.getOffset(), Inc.R.getLength(), ""));
363 auto ErrMsg = "Unexpected conflicts in #include deletions: " +
364 llvm::toString(std::move(Err));
365 llvm_unreachable(ErrMsg.c_str());
371 } // namespace tooling