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/ADT/Optional.h"
14 #include "llvm/Support/FormatVariadic.h"
15 #include "llvm/Support/Path.h"
16
17 namespace clang {
18 namespace tooling {
19 namespace {
20
createLangOpts()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.
getOffsetAfterTokenSequence(StringRef FileName,StringRef Code,const IncludeStyle & Style,llvm::function_ref<unsigned (const SourceManager &,Lexer &,Token &)> GetOffsetAfterSequence)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 Lexer Lex(SM.getMainFileID(), SM.getBufferOrFake(SM.getMainFileID()), SM,
47 createLangOpts());
48 Token Tok;
49 // Get the first token.
50 Lex.LexFromRawLexer(Tok);
51 return GetOffsetAfterSequence(SM, Lex, Tok);
52 }
53
54 // Check if a sequence of tokens is like "#<Name> <raw_identifier>". If it is,
55 // \p Tok will be the token after this directive; otherwise, it can be any token
56 // after the given \p Tok (including \p Tok). If \p RawIDName is provided, the
57 // (second) raw_identifier name is checked.
checkAndConsumeDirectiveWithName(Lexer & Lex,StringRef Name,Token & Tok,llvm::Optional<StringRef> RawIDName=llvm::None)58 bool checkAndConsumeDirectiveWithName(
59 Lexer &Lex, StringRef Name, Token &Tok,
60 llvm::Optional<StringRef> RawIDName = llvm::None) {
61 bool Matched = Tok.is(tok::hash) && !Lex.LexFromRawLexer(Tok) &&
62 Tok.is(tok::raw_identifier) &&
63 Tok.getRawIdentifier() == Name && !Lex.LexFromRawLexer(Tok) &&
64 Tok.is(tok::raw_identifier) &&
65 (!RawIDName || Tok.getRawIdentifier() == *RawIDName);
66 if (Matched)
67 Lex.LexFromRawLexer(Tok);
68 return Matched;
69 }
70
skipComments(Lexer & Lex,Token & Tok)71 void skipComments(Lexer &Lex, Token &Tok) {
72 while (Tok.is(tok::comment))
73 if (Lex.LexFromRawLexer(Tok))
74 return;
75 }
76
77 // Returns the offset after header guard directives and any comments
78 // before/after header guards (e.g. #ifndef/#define pair, #pragma once). If no
79 // header guard is present in the code, this will return the offset after
80 // skipping all comments from the start of the code.
getOffsetAfterHeaderGuardsAndComments(StringRef FileName,StringRef Code,const IncludeStyle & Style)81 unsigned getOffsetAfterHeaderGuardsAndComments(StringRef FileName,
82 StringRef Code,
83 const IncludeStyle &Style) {
84 // \p Consume returns location after header guard or 0 if no header guard is
85 // found.
86 auto ConsumeHeaderGuardAndComment =
87 [&](std::function<unsigned(const SourceManager &SM, Lexer &Lex,
88 Token Tok)>
89 Consume) {
90 return getOffsetAfterTokenSequence(
91 FileName, Code, Style,
92 [&Consume](const SourceManager &SM, Lexer &Lex, Token Tok) {
93 skipComments(Lex, Tok);
94 unsigned InitialOffset = SM.getFileOffset(Tok.getLocation());
95 return std::max(InitialOffset, Consume(SM, Lex, Tok));
96 });
97 };
98 return std::max(
99 // #ifndef/#define
100 ConsumeHeaderGuardAndComment(
101 [](const SourceManager &SM, Lexer &Lex, Token Tok) -> unsigned {
102 if (checkAndConsumeDirectiveWithName(Lex, "ifndef", Tok)) {
103 skipComments(Lex, Tok);
104 if (checkAndConsumeDirectiveWithName(Lex, "define", Tok) &&
105 Tok.isAtStartOfLine())
106 return SM.getFileOffset(Tok.getLocation());
107 }
108 return 0;
109 }),
110 // #pragma once
111 ConsumeHeaderGuardAndComment(
112 [](const SourceManager &SM, Lexer &Lex, Token Tok) -> unsigned {
113 if (checkAndConsumeDirectiveWithName(Lex, "pragma", Tok,
114 StringRef("once")))
115 return SM.getFileOffset(Tok.getLocation());
116 return 0;
117 }));
118 }
119
120 // Check if a sequence of tokens is like
121 // "#include ("header.h" | <header.h>)".
122 // If it is, \p Tok will be the token after this directive; otherwise, it can be
123 // any token after the given \p Tok (including \p Tok).
checkAndConsumeInclusiveDirective(Lexer & Lex,Token & Tok)124 bool checkAndConsumeInclusiveDirective(Lexer &Lex, Token &Tok) {
125 auto Matched = [&]() {
126 Lex.LexFromRawLexer(Tok);
127 return true;
128 };
129 if (Tok.is(tok::hash) && !Lex.LexFromRawLexer(Tok) &&
130 Tok.is(tok::raw_identifier) && Tok.getRawIdentifier() == "include") {
131 if (Lex.LexFromRawLexer(Tok))
132 return false;
133 if (Tok.is(tok::string_literal))
134 return Matched();
135 if (Tok.is(tok::less)) {
136 while (!Lex.LexFromRawLexer(Tok) && Tok.isNot(tok::greater)) {
137 }
138 if (Tok.is(tok::greater))
139 return Matched();
140 }
141 }
142 return false;
143 }
144
145 // Returns the offset of the last #include directive after which a new
146 // #include can be inserted. This ignores #include's after the #include block(s)
147 // in the beginning of a file to avoid inserting headers into code sections
148 // where new #include's should not be added by default.
149 // These code sections include:
150 // - raw string literals (containing #include).
151 // - #if blocks.
152 // - Special #include's among declarations (e.g. functions).
153 //
154 // If no #include after which a new #include can be inserted, this returns the
155 // offset after skipping all comments from the start of the code.
156 // Inserting after an #include is not allowed if it comes after code that is not
157 // #include (e.g. pre-processing directive that is not #include, declarations).
getMaxHeaderInsertionOffset(StringRef FileName,StringRef Code,const IncludeStyle & Style)158 unsigned getMaxHeaderInsertionOffset(StringRef FileName, StringRef Code,
159 const IncludeStyle &Style) {
160 return getOffsetAfterTokenSequence(
161 FileName, Code, Style,
162 [](const SourceManager &SM, Lexer &Lex, Token Tok) {
163 skipComments(Lex, Tok);
164 unsigned MaxOffset = SM.getFileOffset(Tok.getLocation());
165 while (checkAndConsumeInclusiveDirective(Lex, Tok))
166 MaxOffset = SM.getFileOffset(Tok.getLocation());
167 return MaxOffset;
168 });
169 }
170
trimInclude(StringRef IncludeName)171 inline StringRef trimInclude(StringRef IncludeName) {
172 return IncludeName.trim("\"<>");
173 }
174
175 const char IncludeRegexPattern[] =
176 R"(^[\t\ ]*#[\t\ ]*(import|include)[^"<]*(["<][^">]*[">]))";
177
178 // The filename of Path excluding extension.
179 // Used to match implementation with headers, this differs from sys::path::stem:
180 // - in names with multiple dots (foo.cu.cc) it terminates at the *first*
181 // - an empty stem is never returned: /foo/.bar.x => .bar
182 // - we don't bother to handle . and .. specially
matchingStem(llvm::StringRef Path)183 StringRef matchingStem(llvm::StringRef Path) {
184 StringRef Name = llvm::sys::path::filename(Path);
185 return Name.substr(0, Name.find('.', 1));
186 }
187
188 } // anonymous namespace
189
IncludeCategoryManager(const IncludeStyle & Style,StringRef FileName)190 IncludeCategoryManager::IncludeCategoryManager(const IncludeStyle &Style,
191 StringRef FileName)
192 : Style(Style), FileName(FileName) {
193 for (const auto &Category : Style.IncludeCategories) {
194 CategoryRegexs.emplace_back(Category.Regex, Category.RegexIsCaseSensitive
195 ? llvm::Regex::NoFlags
196 : llvm::Regex::IgnoreCase);
197 }
198 IsMainFile = FileName.endswith(".c") || FileName.endswith(".cc") ||
199 FileName.endswith(".cpp") || FileName.endswith(".c++") ||
200 FileName.endswith(".cxx") || FileName.endswith(".m") ||
201 FileName.endswith(".mm");
202 if (!Style.IncludeIsMainSourceRegex.empty()) {
203 llvm::Regex MainFileRegex(Style.IncludeIsMainSourceRegex);
204 IsMainFile |= MainFileRegex.match(FileName);
205 }
206 }
207
getIncludePriority(StringRef IncludeName,bool CheckMainHeader) const208 int IncludeCategoryManager::getIncludePriority(StringRef IncludeName,
209 bool CheckMainHeader) const {
210 int Ret = INT_MAX;
211 for (unsigned i = 0, e = CategoryRegexs.size(); i != e; ++i)
212 if (CategoryRegexs[i].match(IncludeName)) {
213 Ret = Style.IncludeCategories[i].Priority;
214 break;
215 }
216 if (CheckMainHeader && IsMainFile && Ret > 0 && isMainHeader(IncludeName))
217 Ret = 0;
218 return Ret;
219 }
220
getSortIncludePriority(StringRef IncludeName,bool CheckMainHeader) const221 int IncludeCategoryManager::getSortIncludePriority(StringRef IncludeName,
222 bool CheckMainHeader) const {
223 int Ret = INT_MAX;
224 for (unsigned i = 0, e = CategoryRegexs.size(); i != e; ++i)
225 if (CategoryRegexs[i].match(IncludeName)) {
226 Ret = Style.IncludeCategories[i].SortPriority;
227 if (Ret == 0)
228 Ret = Style.IncludeCategories[i].Priority;
229 break;
230 }
231 if (CheckMainHeader && IsMainFile && Ret > 0 && isMainHeader(IncludeName))
232 Ret = 0;
233 return Ret;
234 }
isMainHeader(StringRef IncludeName) const235 bool IncludeCategoryManager::isMainHeader(StringRef IncludeName) const {
236 if (!IncludeName.startswith("\""))
237 return false;
238
239 IncludeName =
240 IncludeName.drop_front(1).drop_back(1); // remove the surrounding "" or <>
241 // Not matchingStem: implementation files may have compound extensions but
242 // headers may not.
243 StringRef HeaderStem = llvm::sys::path::stem(IncludeName);
244 StringRef FileStem = llvm::sys::path::stem(FileName); // foo.cu for foo.cu.cc
245 StringRef MatchingFileStem = matchingStem(FileName); // foo for foo.cu.cc
246 // main-header examples:
247 // 1) foo.h => foo.cc
248 // 2) foo.h => foo.cu.cc
249 // 3) foo.proto.h => foo.proto.cc
250 //
251 // non-main-header examples:
252 // 1) foo.h => bar.cc
253 // 2) foo.proto.h => foo.cc
254 StringRef Matching;
255 if (MatchingFileStem.startswith_lower(HeaderStem))
256 Matching = MatchingFileStem; // example 1), 2)
257 else if (FileStem.equals_lower(HeaderStem))
258 Matching = FileStem; // example 3)
259 if (!Matching.empty()) {
260 llvm::Regex MainIncludeRegex(HeaderStem.str() + Style.IncludeIsMainRegex,
261 llvm::Regex::IgnoreCase);
262 if (MainIncludeRegex.match(Matching))
263 return true;
264 }
265 return false;
266 }
267
HeaderIncludes(StringRef FileName,StringRef Code,const IncludeStyle & Style)268 HeaderIncludes::HeaderIncludes(StringRef FileName, StringRef Code,
269 const IncludeStyle &Style)
270 : FileName(FileName), Code(Code), FirstIncludeOffset(-1),
271 MinInsertOffset(
272 getOffsetAfterHeaderGuardsAndComments(FileName, Code, Style)),
273 MaxInsertOffset(MinInsertOffset +
274 getMaxHeaderInsertionOffset(
275 FileName, Code.drop_front(MinInsertOffset), Style)),
276 Categories(Style, FileName),
277 IncludeRegex(llvm::Regex(IncludeRegexPattern)) {
278 // Add 0 for main header and INT_MAX for headers that are not in any
279 // category.
280 Priorities = {0, INT_MAX};
281 for (const auto &Category : Style.IncludeCategories)
282 Priorities.insert(Category.Priority);
283 SmallVector<StringRef, 32> Lines;
284 Code.drop_front(MinInsertOffset).split(Lines, "\n");
285
286 unsigned Offset = MinInsertOffset;
287 unsigned NextLineOffset;
288 SmallVector<StringRef, 4> Matches;
289 for (auto Line : Lines) {
290 NextLineOffset = std::min(Code.size(), Offset + Line.size() + 1);
291 if (IncludeRegex.match(Line, &Matches)) {
292 // If this is the last line without trailing newline, we need to make
293 // sure we don't delete across the file boundary.
294 addExistingInclude(
295 Include(Matches[2],
296 tooling::Range(
297 Offset, std::min(Line.size() + 1, Code.size() - Offset))),
298 NextLineOffset);
299 }
300 Offset = NextLineOffset;
301 }
302
303 // Populate CategoryEndOfssets:
304 // - Ensure that CategoryEndOffset[Highest] is always populated.
305 // - If CategoryEndOffset[Priority] isn't set, use the next higher value
306 // that is set, up to CategoryEndOffset[Highest].
307 auto Highest = Priorities.begin();
308 if (CategoryEndOffsets.find(*Highest) == CategoryEndOffsets.end()) {
309 if (FirstIncludeOffset >= 0)
310 CategoryEndOffsets[*Highest] = FirstIncludeOffset;
311 else
312 CategoryEndOffsets[*Highest] = MinInsertOffset;
313 }
314 // By this point, CategoryEndOffset[Highest] is always set appropriately:
315 // - to an appropriate location before/after existing #includes, or
316 // - to right after the header guard, or
317 // - to the beginning of the file.
318 for (auto I = ++Priorities.begin(), E = Priorities.end(); I != E; ++I)
319 if (CategoryEndOffsets.find(*I) == CategoryEndOffsets.end())
320 CategoryEndOffsets[*I] = CategoryEndOffsets[*std::prev(I)];
321 }
322
323 // \p Offset: the start of the line following this include directive.
addExistingInclude(Include IncludeToAdd,unsigned NextLineOffset)324 void HeaderIncludes::addExistingInclude(Include IncludeToAdd,
325 unsigned NextLineOffset) {
326 auto Iter =
327 ExistingIncludes.try_emplace(trimInclude(IncludeToAdd.Name)).first;
328 Iter->second.push_back(std::move(IncludeToAdd));
329 auto &CurInclude = Iter->second.back();
330 // The header name with quotes or angle brackets.
331 // Only record the offset of current #include if we can insert after it.
332 if (CurInclude.R.getOffset() <= MaxInsertOffset) {
333 int Priority = Categories.getIncludePriority(
334 CurInclude.Name, /*CheckMainHeader=*/FirstIncludeOffset < 0);
335 CategoryEndOffsets[Priority] = NextLineOffset;
336 IncludesByPriority[Priority].push_back(&CurInclude);
337 if (FirstIncludeOffset < 0)
338 FirstIncludeOffset = CurInclude.R.getOffset();
339 }
340 }
341
342 llvm::Optional<tooling::Replacement>
insert(llvm::StringRef IncludeName,bool IsAngled) const343 HeaderIncludes::insert(llvm::StringRef IncludeName, bool IsAngled) const {
344 assert(IncludeName == trimInclude(IncludeName));
345 // If a <header> ("header") already exists in code, "header" (<header>) with
346 // different quotation will still be inserted.
347 // FIXME: figure out if this is the best behavior.
348 auto It = ExistingIncludes.find(IncludeName);
349 if (It != ExistingIncludes.end())
350 for (const auto &Inc : It->second)
351 if ((IsAngled && StringRef(Inc.Name).startswith("<")) ||
352 (!IsAngled && StringRef(Inc.Name).startswith("\"")))
353 return llvm::None;
354 std::string Quoted =
355 std::string(llvm::formatv(IsAngled ? "<{0}>" : "\"{0}\"", IncludeName));
356 StringRef QuotedName = Quoted;
357 int Priority = Categories.getIncludePriority(
358 QuotedName, /*CheckMainHeader=*/FirstIncludeOffset < 0);
359 auto CatOffset = CategoryEndOffsets.find(Priority);
360 assert(CatOffset != CategoryEndOffsets.end());
361 unsigned InsertOffset = CatOffset->second; // Fall back offset
362 auto Iter = IncludesByPriority.find(Priority);
363 if (Iter != IncludesByPriority.end()) {
364 for (const auto *Inc : Iter->second) {
365 if (QuotedName < Inc->Name) {
366 InsertOffset = Inc->R.getOffset();
367 break;
368 }
369 }
370 }
371 assert(InsertOffset <= Code.size());
372 std::string NewInclude =
373 std::string(llvm::formatv("#include {0}\n", QuotedName));
374 // When inserting headers at end of the code, also append '\n' to the code
375 // if it does not end with '\n'.
376 // FIXME: when inserting multiple #includes at the end of code, only one
377 // newline should be added.
378 if (InsertOffset == Code.size() && (!Code.empty() && Code.back() != '\n'))
379 NewInclude = "\n" + NewInclude;
380 return tooling::Replacement(FileName, InsertOffset, 0, NewInclude);
381 }
382
remove(llvm::StringRef IncludeName,bool IsAngled) const383 tooling::Replacements HeaderIncludes::remove(llvm::StringRef IncludeName,
384 bool IsAngled) const {
385 assert(IncludeName == trimInclude(IncludeName));
386 tooling::Replacements Result;
387 auto Iter = ExistingIncludes.find(IncludeName);
388 if (Iter == ExistingIncludes.end())
389 return Result;
390 for (const auto &Inc : Iter->second) {
391 if ((IsAngled && StringRef(Inc.Name).startswith("\"")) ||
392 (!IsAngled && StringRef(Inc.Name).startswith("<")))
393 continue;
394 llvm::Error Err = Result.add(tooling::Replacement(
395 FileName, Inc.R.getOffset(), Inc.R.getLength(), ""));
396 if (Err) {
397 auto ErrMsg = "Unexpected conflicts in #include deletions: " +
398 llvm::toString(std::move(Err));
399 llvm_unreachable(ErrMsg.c_str());
400 }
401 }
402 return Result;
403 }
404
405 } // namespace tooling
406 } // namespace clang
407