1 //===--- IncludeCleaner.cpp - Unused/Missing Headers Analysis ---*- 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 "IncludeCleaner.h"
10 #include "Diagnostics.h"
12 #include "ParsedAST.h"
15 #include "SourceCode.h"
16 #include "clang-include-cleaner/Analysis.h"
17 #include "clang-include-cleaner/IncludeSpeller.h"
18 #include "clang-include-cleaner/Record.h"
19 #include "clang-include-cleaner/Types.h"
20 #include "support/Logger.h"
21 #include "support/Path.h"
22 #include "support/Trace.h"
23 #include "clang/AST/ASTContext.h"
24 #include "clang/Basic/Diagnostic.h"
25 #include "clang/Basic/LLVM.h"
26 #include "clang/Basic/SourceLocation.h"
27 #include "clang/Basic/SourceManager.h"
28 #include "clang/Format/Format.h"
29 #include "clang/Lex/DirectoryLookup.h"
30 #include "clang/Lex/HeaderSearch.h"
31 #include "clang/Lex/Preprocessor.h"
32 #include "clang/Tooling/Core/Replacement.h"
33 #include "clang/Tooling/Inclusions/HeaderIncludes.h"
34 #include "clang/Tooling/Inclusions/StandardLibrary.h"
35 #include "clang/Tooling/Syntax/Tokens.h"
36 #include "llvm/ADT/ArrayRef.h"
37 #include "llvm/ADT/DenseSet.h"
38 #include "llvm/ADT/GenericUniformityImpl.h"
39 #include "llvm/ADT/STLExtras.h"
40 #include "llvm/ADT/SmallString.h"
41 #include "llvm/ADT/SmallVector.h"
42 #include "llvm/ADT/StringRef.h"
43 #include "llvm/Support/Error.h"
44 #include "llvm/Support/ErrorHandling.h"
45 #include "llvm/Support/FormatVariadic.h"
46 #include "llvm/Support/Path.h"
47 #include "llvm/Support/Regex.h"
57 namespace clang::clangd
{
60 bool isIgnored(llvm::StringRef HeaderPath
, HeaderFilter IgnoreHeaders
) {
61 // Convert the path to Unix slashes and try to match against the filter.
62 llvm::SmallString
<64> NormalizedPath(HeaderPath
);
63 llvm::sys::path::native(NormalizedPath
, llvm::sys::path::Style::posix
);
64 for (auto &Filter
: IgnoreHeaders
) {
65 if (Filter(NormalizedPath
))
71 bool mayConsiderUnused(const Inclusion
&Inc
, ParsedAST
&AST
,
72 const include_cleaner::PragmaIncludes
*PI
) {
74 auto HID
= static_cast<IncludeStructure::HeaderID
>(*Inc
.HeaderID
);
75 auto FE
= AST
.getSourceManager().getFileManager().getFileRef(
76 AST
.getIncludeStructure().getRealPath(HID
));
78 if (PI
&& PI
->shouldKeep(*FE
))
80 // FIXME(kirillbobyrev): We currently do not support the umbrella headers.
81 // System headers are likely to be standard library headers.
82 // Until we have good support for umbrella headers, don't warn about them.
83 if (Inc
.Written
.front() == '<')
84 return tooling::stdlib::Header::named(Inc
.Written
).has_value();
86 // Check if main file is the public interface for a private header. If so we
87 // shouldn't diagnose it as unused.
88 if (auto PHeader
= PI
->getPublic(*FE
); !PHeader
.empty()) {
89 PHeader
= PHeader
.trim("<>\"");
90 // Since most private -> public mappings happen in a verbatim way, we
91 // check textually here. This might go wrong in presence of symlinks or
92 // header mappings. But that's not different than rest of the places.
93 if (AST
.tuPath().endswith(PHeader
))
97 // Headers without include guards have side effects and are not
98 // self-contained, skip them.
99 if (!AST
.getPreprocessor().getHeaderSearchInfo().isFileMultipleIncludeGuarded(
100 &FE
->getFileEntry())) {
101 dlog("{0} doesn't have header guard and will not be considered unused",
108 std::vector
<Diag
> generateMissingIncludeDiagnostics(
109 ParsedAST
&AST
, llvm::ArrayRef
<MissingIncludeDiagInfo
> MissingIncludes
,
110 llvm::StringRef Code
, HeaderFilter IgnoreHeaders
) {
111 std::vector
<Diag
> Result
;
112 const SourceManager
&SM
= AST
.getSourceManager();
113 const FileEntry
*MainFile
= SM
.getFileEntryForID(SM
.getMainFileID());
115 auto FileStyle
= format::getStyle(
116 format::DefaultFormatStyle
, AST
.tuPath(), format::DefaultFallbackStyle
,
117 Code
, &SM
.getFileManager().getVirtualFileSystem());
119 elog("Couldn't infer style", FileStyle
.takeError());
120 FileStyle
= format::getLLVMStyle();
123 tooling::HeaderIncludes
HeaderIncludes(AST
.tuPath(), Code
,
124 FileStyle
->IncludeStyle
);
125 for (const auto &SymbolWithMissingInclude
: MissingIncludes
) {
126 llvm::StringRef ResolvedPath
=
127 SymbolWithMissingInclude
.Providers
.front().resolvedPath();
128 if (isIgnored(ResolvedPath
, IgnoreHeaders
)) {
129 dlog("IncludeCleaner: not diagnosing missing include {0}, filtered by "
135 std::string Spelling
= include_cleaner::spellHeader(
136 {SymbolWithMissingInclude
.Providers
.front(),
137 AST
.getPreprocessor().getHeaderSearchInfo(), MainFile
});
139 llvm::StringRef HeaderRef
{Spelling
};
140 bool Angled
= HeaderRef
.starts_with("<");
141 // We might suggest insertion of an existing include in edge cases, e.g.,
142 // include is present in a PP-disabled region, or spelling of the header
143 // turns out to be the same as one of the unresolved includes in the
145 std::optional
<tooling::Replacement
> Replacement
= HeaderIncludes
.insert(
146 HeaderRef
.trim("\"<>"), Angled
, tooling::IncludeDirective::Include
);
147 if (!Replacement
.has_value())
150 Diag
&D
= Result
.emplace_back();
152 llvm::formatv("No header providing \"{0}\" is directly included",
153 SymbolWithMissingInclude
.Symbol
.name());
154 D
.Name
= "missing-includes";
155 D
.Source
= Diag::DiagSource::Clangd
;
156 D
.File
= AST
.tuPath();
157 D
.InsideMainFile
= true;
158 // We avoid the "warning" severity here in favor of LSP's "information".
160 // Users treat most warnings on code being edited as high-priority.
161 // They don't think of include cleanups the same way: they want to edit
162 // lines with existing violations without fixing them.
163 // Diagnostics at the same level tend to be visually indistinguishable,
164 // and a few missing includes can cause many diagnostics.
165 // Marking these as "information" leaves them visible, but less intrusive.
167 // (These concerns don't apply to unused #include warnings: these are fewer,
168 // they appear on infrequently-edited lines with few other warnings, and
169 // the 'Unneccesary' tag often result in a different rendering)
171 // Usually clang's "note" severity usually has special semantics, being
172 // translated into LSP RelatedInformation of a parent diagnostic.
173 // But not here: these aren't processed by clangd's DiagnosticConsumer.
174 D
.Severity
= DiagnosticsEngine::Note
;
175 D
.Range
= clangd::Range
{
176 offsetToPosition(Code
,
177 SymbolWithMissingInclude
.SymRefRange
.beginOffset()),
178 offsetToPosition(Code
,
179 SymbolWithMissingInclude
.SymRefRange
.endOffset())};
180 auto &F
= D
.Fixes
.emplace_back();
181 F
.Message
= "#include " + Spelling
;
182 TextEdit Edit
= replacementToEdit(Code
, *Replacement
);
183 F
.Edits
.emplace_back(std::move(Edit
));
188 std::vector
<Diag
> generateUnusedIncludeDiagnostics(
189 PathRef FileName
, llvm::ArrayRef
<const Inclusion
*> UnusedIncludes
,
190 llvm::StringRef Code
, HeaderFilter IgnoreHeaders
) {
191 std::vector
<Diag
> Result
;
192 for (const auto *Inc
: UnusedIncludes
) {
193 if (isIgnored(Inc
->Resolved
, IgnoreHeaders
))
195 Diag
&D
= Result
.emplace_back();
197 llvm::formatv("included header {0} is not used directly",
198 llvm::sys::path::filename(
199 Inc
->Written
.substr(1, Inc
->Written
.size() - 2),
200 llvm::sys::path::Style::posix
));
201 D
.Name
= "unused-includes";
202 D
.Source
= Diag::DiagSource::Clangd
;
204 D
.InsideMainFile
= true;
205 D
.Severity
= DiagnosticsEngine::Warning
;
206 D
.Tags
.push_back(Unnecessary
);
207 D
.Range
= rangeTillEOL(Code
, Inc
->HashOffset
);
208 // FIXME(kirillbobyrev): Removing inclusion might break the code if the
209 // used headers are only reachable transitively through this one. Suggest
210 // including them directly instead.
211 // FIXME(kirillbobyrev): Add fix suggestion for adding IWYU pragmas
212 // (keep/export) remove the warning once we support IWYU pragmas.
213 auto &F
= D
.Fixes
.emplace_back();
214 F
.Message
= "remove #include directive";
215 F
.Edits
.emplace_back();
216 F
.Edits
.back().range
.start
.line
= Inc
->HashLine
;
217 F
.Edits
.back().range
.end
.line
= Inc
->HashLine
+ 1;
223 removeAllUnusedIncludes(llvm::ArrayRef
<Diag
> UnusedIncludes
) {
224 if (UnusedIncludes
.empty())
228 RemoveAll
.Message
= "remove all unused includes";
229 for (const auto &Diag
: UnusedIncludes
) {
230 assert(Diag
.Fixes
.size() == 1 && "Expected exactly one fix.");
231 RemoveAll
.Edits
.insert(RemoveAll
.Edits
.end(),
232 Diag
.Fixes
.front().Edits
.begin(),
233 Diag
.Fixes
.front().Edits
.end());
236 // TODO(hokein): emit a suitable text for the label.
237 ChangeAnnotation Annotation
= {/*label=*/"",
238 /*needsConfirmation=*/true,
240 static const ChangeAnnotationIdentifier RemoveAllUnusedID
=
241 "RemoveAllUnusedIncludes";
242 for (unsigned I
= 0; I
< RemoveAll
.Edits
.size(); ++I
) {
243 ChangeAnnotationIdentifier ID
= RemoveAllUnusedID
+ std::to_string(I
);
244 RemoveAll
.Edits
[I
].annotationId
= ID
;
245 RemoveAll
.Annotations
.push_back({ID
, Annotation
});
251 addAllMissingIncludes(llvm::ArrayRef
<Diag
> MissingIncludeDiags
) {
252 if (MissingIncludeDiags
.empty())
256 AddAllMissing
.Message
= "add all missing includes";
257 // A map to deduplicate the edits with the same new text.
258 // newText (#include "my_missing_header.h") -> TextEdit.
259 std::map
<std::string
, TextEdit
> Edits
;
260 for (const auto &Diag
: MissingIncludeDiags
) {
261 assert(Diag
.Fixes
.size() == 1 && "Expected exactly one fix.");
262 for (const auto &Edit
: Diag
.Fixes
.front().Edits
) {
263 Edits
.try_emplace(Edit
.newText
, Edit
);
266 // FIXME(hokein): emit used symbol reference in the annotation.
267 ChangeAnnotation Annotation
= {/*label=*/"",
268 /*needsConfirmation=*/true,
270 static const ChangeAnnotationIdentifier AddAllMissingID
=
271 "AddAllMissingIncludes";
273 for (auto &It
: Edits
) {
274 ChangeAnnotationIdentifier ID
= AddAllMissingID
+ std::to_string(I
++);
275 AddAllMissing
.Edits
.push_back(std::move(It
.second
));
276 AddAllMissing
.Edits
.back().annotationId
= ID
;
278 AddAllMissing
.Annotations
.push_back({ID
, Annotation
});
280 return AddAllMissing
;
282 Fix
fixAll(const Fix
&RemoveAllUnused
, const Fix
&AddAllMissing
) {
284 FixAll
.Message
= "fix all includes";
286 for (const auto &F
: RemoveAllUnused
.Edits
)
287 FixAll
.Edits
.push_back(F
);
288 for (const auto &F
: AddAllMissing
.Edits
)
289 FixAll
.Edits
.push_back(F
);
291 for (const auto &A
: RemoveAllUnused
.Annotations
)
292 FixAll
.Annotations
.push_back(A
);
293 for (const auto &A
: AddAllMissing
.Annotations
)
294 FixAll
.Annotations
.push_back(A
);
298 std::vector
<const Inclusion
*>
299 getUnused(ParsedAST
&AST
,
300 const llvm::DenseSet
<IncludeStructure::HeaderID
> &ReferencedFiles
) {
301 trace::Span
Tracer("IncludeCleaner::getUnused");
302 std::vector
<const Inclusion
*> Unused
;
303 for (const Inclusion
&MFI
: AST
.getIncludeStructure().MainFileIncludes
) {
306 auto IncludeID
= static_cast<IncludeStructure::HeaderID
>(*MFI
.HeaderID
);
307 if (ReferencedFiles
.contains(IncludeID
))
309 if (!mayConsiderUnused(MFI
, AST
, AST
.getPragmaIncludes().get())) {
310 dlog("{0} was not used, but is not eligible to be diagnosed as unused",
314 Unused
.push_back(&MFI
);
321 std::vector
<include_cleaner::SymbolReference
>
322 collectMacroReferences(ParsedAST
&AST
) {
323 const auto &SM
= AST
.getSourceManager();
324 auto &PP
= AST
.getPreprocessor();
325 std::vector
<include_cleaner::SymbolReference
> Macros
;
326 for (const auto &[_
, Refs
] : AST
.getMacros().MacroRefs
) {
327 for (const auto &Ref
: Refs
) {
328 auto Loc
= SM
.getComposedLoc(SM
.getMainFileID(), Ref
.StartOffset
);
329 const auto *Tok
= AST
.getTokens().spelledTokenAt(Loc
);
332 auto Macro
= locateMacroAt(*Tok
, PP
);
335 auto DefLoc
= Macro
->NameLoc
;
336 if (!DefLoc
.isValid())
339 {include_cleaner::Macro
{/*Name=*/PP
.getIdentifierInfo(Tok
->text(SM
)),
342 Ref
.InConditionalDirective
? include_cleaner::RefType::Ambiguous
343 : include_cleaner::RefType::Explicit
});
350 include_cleaner::Includes
convertIncludes(const ParsedAST
&AST
) {
351 auto &SM
= AST
.getSourceManager();
353 include_cleaner::Includes ConvertedIncludes
;
354 // We satisfy Includes's contract that search dirs and included files have
355 // matching path styles: both ultimately use FileManager::getCanonicalName().
356 for (const auto &Dir
: AST
.getIncludeStructure().SearchPathsCanonical
)
357 ConvertedIncludes
.addSearchDirectory(Dir
);
359 for (const Inclusion
&Inc
: AST
.getIncludeStructure().MainFileIncludes
) {
360 include_cleaner::Include TransformedInc
;
361 llvm::StringRef WrittenRef
= llvm::StringRef(Inc
.Written
);
362 TransformedInc
.Spelled
= WrittenRef
.trim("\"<>");
363 TransformedInc
.HashLocation
=
364 SM
.getComposedLoc(SM
.getMainFileID(), Inc
.HashOffset
);
365 TransformedInc
.Line
= Inc
.HashLine
+ 1;
366 TransformedInc
.Angled
= WrittenRef
.starts_with("<");
367 // Inc.Resolved is canonicalized with clangd::getCanonicalPath(),
368 // which is based on FileManager::getCanonicalName(ParentDir).
369 auto FE
= SM
.getFileManager().getFileRef(Inc
.Resolved
);
371 elog("IncludeCleaner: Failed to get an entry for resolved path {0}: {1}",
372 Inc
.Resolved
, FE
.takeError());
375 TransformedInc
.Resolved
= *FE
;
376 ConvertedIncludes
.add(std::move(TransformedInc
));
378 return ConvertedIncludes
;
381 IncludeCleanerFindings
computeIncludeCleanerFindings(ParsedAST
&AST
) {
382 // Interaction is only polished for C/CPP.
383 if (AST
.getLangOpts().ObjC
)
385 const auto &SM
= AST
.getSourceManager();
386 include_cleaner::Includes ConvertedIncludes
= convertIncludes(AST
);
387 const FileEntry
*MainFile
= SM
.getFileEntryForID(SM
.getMainFileID());
388 auto *PreamblePatch
= PreamblePatch::getPatchEntry(AST
.tuPath(), SM
);
390 std::vector
<include_cleaner::SymbolReference
> Macros
=
391 collectMacroReferences(AST
);
392 std::vector
<MissingIncludeDiagInfo
> MissingIncludes
;
393 llvm::DenseSet
<IncludeStructure::HeaderID
> Used
;
394 trace::Span
Tracer("include_cleaner::walkUsed");
395 include_cleaner::walkUsed(
396 AST
.getLocalTopLevelDecls(), /*MacroRefs=*/Macros
,
397 AST
.getPragmaIncludes().get(), AST
.getPreprocessor(),
398 [&](const include_cleaner::SymbolReference
&Ref
,
399 llvm::ArrayRef
<include_cleaner::Header
> Providers
) {
400 bool Satisfied
= false;
401 for (const auto &H
: Providers
) {
402 if (H
.kind() == include_cleaner::Header::Physical
&&
403 (H
.physical() == MainFile
|| H
.physical() == PreamblePatch
)) {
407 for (auto *Inc
: ConvertedIncludes
.match(H
)) {
410 AST
.getIncludeStructure().getID(&Inc
->Resolved
->getFileEntry());
411 assert(HeaderID
.has_value() &&
412 "ConvertedIncludes only contains resolved includes.");
413 Used
.insert(*HeaderID
);
417 if (Satisfied
|| Providers
.empty() ||
418 Ref
.RT
!= include_cleaner::RefType::Explicit
)
421 // We actually always want to map usages to their spellings, but
422 // spelling locations can point into preamble section. Using these
423 // offsets could lead into crashes in presence of stale preambles. Hence
424 // we use "getFileLoc" instead to make sure it always points into main
426 // FIXME: Use presumed locations to map such usages back to patched
428 auto Loc
= SM
.getFileLoc(Ref
.RefLocation
);
429 // File locations can be outside of the main file if macro is expanded
430 // through an #include.
431 while (SM
.getFileID(Loc
) != SM
.getMainFileID())
432 Loc
= SM
.getIncludeLoc(SM
.getFileID(Loc
));
433 auto TouchingTokens
=
434 syntax::spelledTokensTouching(Loc
, AST
.getTokens());
435 assert(!TouchingTokens
.empty());
436 // Loc points to the start offset of the ref token, here we use the last
437 // element of the TouchingTokens, e.g. avoid getting the "::" for
439 MissingIncludeDiagInfo DiagInfo
{
440 Ref
.Target
, TouchingTokens
.back().range(SM
), Providers
};
441 MissingIncludes
.push_back(std::move(DiagInfo
));
443 // Put possibly equal diagnostics together for deduplication.
444 // The duplicates might be from macro arguments that get expanded multiple
446 llvm::stable_sort(MissingIncludes
, [](const MissingIncludeDiagInfo
&LHS
,
447 const MissingIncludeDiagInfo
&RHS
) {
448 // First sort by reference location.
449 if (LHS
.SymRefRange
!= RHS
.SymRefRange
) {
450 // We can get away just by comparing the offsets as all the ranges are in
452 return LHS
.SymRefRange
.beginOffset() < RHS
.SymRefRange
.beginOffset();
454 // For the same location, break ties using the symbol. Note that this won't
455 // be stable across runs.
456 using MapInfo
= llvm::DenseMapInfo
<include_cleaner::Symbol
>;
457 return MapInfo::getHashValue(LHS
.Symbol
) <
458 MapInfo::getHashValue(RHS
.Symbol
);
460 MissingIncludes
.erase(llvm::unique(MissingIncludes
), MissingIncludes
.end());
461 std::vector
<const Inclusion
*> UnusedIncludes
= getUnused(AST
, Used
);
462 return {std::move(UnusedIncludes
), std::move(MissingIncludes
)};
465 bool isPreferredProvider(const Inclusion
&Inc
,
466 const include_cleaner::Includes
&Includes
,
467 llvm::ArrayRef
<include_cleaner::Header
> Providers
) {
468 for (const auto &H
: Providers
) {
469 auto Matches
= Includes
.match(H
);
470 for (const include_cleaner::Include
*Match
: Matches
)
471 if (Match
->Line
== unsigned(Inc
.HashLine
+ 1))
472 return true; // this header is (equal) best
473 if (!Matches
.empty())
474 return false; // another header is better
476 return false; // no header provides the symbol
480 issueIncludeCleanerDiagnostics(ParsedAST
&AST
, llvm::StringRef Code
,
481 const IncludeCleanerFindings
&Findings
,
482 HeaderFilter IgnoreHeaders
) {
483 trace::Span
Tracer("IncludeCleaner::issueIncludeCleanerDiagnostics");
484 std::vector
<Diag
> UnusedIncludes
= generateUnusedIncludeDiagnostics(
485 AST
.tuPath(), Findings
.UnusedIncludes
, Code
, IgnoreHeaders
);
486 std::optional
<Fix
> RemoveAllUnused
= removeAllUnusedIncludes(UnusedIncludes
);
488 std::vector
<Diag
> MissingIncludeDiags
= generateMissingIncludeDiagnostics(
489 AST
, Findings
.MissingIncludes
, Code
, IgnoreHeaders
);
490 std::optional
<Fix
> AddAllMissing
= addAllMissingIncludes(MissingIncludeDiags
);
492 std::optional
<Fix
> FixAll
;
493 if (RemoveAllUnused
&& AddAllMissing
)
494 FixAll
= fixAll(*RemoveAllUnused
, *AddAllMissing
);
496 auto AddBatchFix
= [](const std::optional
<Fix
> &F
, clang::clangd::Diag
*Out
) {
499 Out
->Fixes
.push_back(*F
);
501 for (auto &Diag
: MissingIncludeDiags
) {
502 AddBatchFix(MissingIncludeDiags
.size() > 1 ? AddAllMissing
: std::nullopt
,
504 AddBatchFix(FixAll
, &Diag
);
506 for (auto &Diag
: UnusedIncludes
) {
507 AddBatchFix(UnusedIncludes
.size() > 1 ? RemoveAllUnused
: std::nullopt
,
509 AddBatchFix(FixAll
, &Diag
);
512 auto Result
= std::move(MissingIncludeDiags
);
513 llvm::move(UnusedIncludes
, std::back_inserter(Result
));
517 } // namespace clang::clangd