1 //===--- IncludeFixer.cpp ----------------------------------------*- 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 "IncludeFixer.h"
10 #include "AST.h"
11 #include "Diagnostics.h"
12 #include "SourceCode.h"
13 #include "index/Index.h"
14 #include "index/Symbol.h"
15 #include "support/Logger.h"
16 #include "support/Trace.h"
17 #include "clang/AST/Decl.h"
18 #include "clang/AST/DeclBase.h"
19 #include "clang/AST/DeclarationName.h"
20 #include "clang/AST/NestedNameSpecifier.h"
21 #include "clang/AST/Type.h"
22 #include "clang/Basic/Diagnostic.h"
23 #include "clang/Basic/DiagnosticSema.h"
24 #include "clang/Basic/LangOptions.h"
25 #include "clang/Basic/SourceLocation.h"
26 #include "clang/Basic/SourceManager.h"
27 #include "clang/Basic/TokenKinds.h"
28 #include "clang/Lex/Lexer.h"
29 #include "clang/Sema/DeclSpec.h"
30 #include "clang/Sema/Lookup.h"
31 #include "clang/Sema/Scope.h"
32 #include "clang/Sema/Sema.h"
33 #include "clang/Sema/TypoCorrection.h"
34 #include "llvm/ADT/ArrayRef.h"
35 #include "llvm/ADT/DenseMap.h"
36 #include "llvm/ADT/None.h"
37 #include "llvm/ADT/Optional.h"
38 #include "llvm/ADT/StringExtras.h"
39 #include "llvm/ADT/StringRef.h"
40 #include "llvm/ADT/StringSet.h"
41 #include "llvm/Support/Error.h"
42 #include "llvm/Support/FormatVariadic.h"
43 #include <algorithm>
44 #include <set>
45 #include <string>
46 #include <vector>
47
48 namespace clang {
49 namespace clangd {
50
fix(DiagnosticsEngine::Level DiagLevel,const clang::Diagnostic & Info) const51 std::vector<Fix> IncludeFixer::fix(DiagnosticsEngine::Level DiagLevel,
52 const clang::Diagnostic &Info) const {
53 switch (Info.getID()) {
54 case diag::err_incomplete_nested_name_spec:
55 case diag::err_incomplete_base_class:
56 case diag::err_incomplete_member_access:
57 case diag::err_incomplete_type:
58 case diag::err_typecheck_decl_incomplete_type:
59 case diag::err_typecheck_incomplete_tag:
60 case diag::err_invalid_incomplete_type_use:
61 case diag::err_sizeof_alignof_incomplete_or_sizeless_type:
62 case diag::err_for_range_incomplete_type:
63 case diag::err_func_def_incomplete_result:
64 case diag::err_field_incomplete_or_sizeless:
65 // Incomplete type diagnostics should have a QualType argument for the
66 // incomplete type.
67 for (unsigned Idx = 0; Idx < Info.getNumArgs(); ++Idx) {
68 if (Info.getArgKind(Idx) == DiagnosticsEngine::ak_qualtype) {
69 auto QT = QualType::getFromOpaquePtr((void *)Info.getRawArg(Idx));
70 if (const Type *T = QT.getTypePtrOrNull())
71 if (T->isIncompleteType())
72 return fixIncompleteType(*T);
73 }
74 }
75 break;
76 case diag::err_unknown_typename:
77 case diag::err_unknown_typename_suggest:
78 case diag::err_typename_nested_not_found:
79 case diag::err_no_template:
80 case diag::err_no_template_suggest:
81 case diag::err_undeclared_use:
82 case diag::err_undeclared_use_suggest:
83 case diag::err_undeclared_var_use:
84 case diag::err_undeclared_var_use_suggest:
85 case diag::err_no_member: // Could be no member in namespace.
86 case diag::err_no_member_suggest:
87 case diag::err_no_member_template:
88 case diag::err_no_member_template_suggest:
89 if (LastUnresolvedName) {
90 // Try to fix unresolved name caused by missing declaration.
91 // E.g.
92 // clang::SourceManager SM;
93 // ~~~~~~~~~~~~~
94 // UnresolvedName
95 // or
96 // namespace clang { SourceManager SM; }
97 // ~~~~~~~~~~~~~
98 // UnresolvedName
99 // We only attempt to recover a diagnostic if it has the same location as
100 // the last seen unresolved name.
101 if (DiagLevel >= DiagnosticsEngine::Error &&
102 LastUnresolvedName->Loc == Info.getLocation())
103 return fixUnresolvedName();
104 }
105 }
106 return {};
107 }
108
fixIncompleteType(const Type & T) const109 std::vector<Fix> IncludeFixer::fixIncompleteType(const Type &T) const {
110 // Only handle incomplete TagDecl type.
111 const TagDecl *TD = T.getAsTagDecl();
112 if (!TD)
113 return {};
114 std::string TypeName = printQualifiedName(*TD);
115 trace::Span Tracer("Fix include for incomplete type");
116 SPAN_ATTACH(Tracer, "type", TypeName);
117 vlog("Trying to fix include for incomplete type {0}", TypeName);
118
119 auto ID = getSymbolID(TD);
120 if (!ID)
121 return {};
122 llvm::Optional<const SymbolSlab *> Symbols = lookupCached(ID);
123 if (!Symbols)
124 return {};
125 const SymbolSlab &Syms = **Symbols;
126 std::vector<Fix> Fixes;
127 if (!Syms.empty()) {
128 auto &Matched = *Syms.begin();
129 if (!Matched.IncludeHeaders.empty() && Matched.Definition &&
130 Matched.CanonicalDeclaration.FileURI == Matched.Definition.FileURI)
131 Fixes = fixesForSymbols(Syms);
132 }
133 return Fixes;
134 }
135
fixesForSymbols(const SymbolSlab & Syms) const136 std::vector<Fix> IncludeFixer::fixesForSymbols(const SymbolSlab &Syms) const {
137 auto Inserted = [&](const Symbol &Sym, llvm::StringRef Header)
138 -> llvm::Expected<std::pair<std::string, bool>> {
139 auto ResolvedDeclaring =
140 URI::resolve(Sym.CanonicalDeclaration.FileURI, File);
141 if (!ResolvedDeclaring)
142 return ResolvedDeclaring.takeError();
143 auto ResolvedInserted = toHeaderFile(Header, File);
144 if (!ResolvedInserted)
145 return ResolvedInserted.takeError();
146 auto Spelled = Inserter->calculateIncludePath(*ResolvedInserted, File);
147 if (!Spelled)
148 return error("Header not on include path");
149 return std::make_pair(
150 std::move(*Spelled),
151 Inserter->shouldInsertInclude(*ResolvedDeclaring, *ResolvedInserted));
152 };
153
154 std::vector<Fix> Fixes;
155 // Deduplicate fixes by include headers. This doesn't distinguish symbols in
156 // different scopes from the same header, but this case should be rare and is
157 // thus ignored.
158 llvm::StringSet<> InsertedHeaders;
159 for (const auto &Sym : Syms) {
160 for (const auto &Inc : getRankedIncludes(Sym)) {
161 if (auto ToInclude = Inserted(Sym, Inc)) {
162 if (ToInclude->second) {
163 auto I = InsertedHeaders.try_emplace(ToInclude->first);
164 if (!I.second)
165 continue;
166 if (auto Edit = Inserter->insert(ToInclude->first))
167 Fixes.push_back(Fix{std::string(llvm::formatv(
168 "Add include {0} for symbol {1}{2}",
169 ToInclude->first, Sym.Scope, Sym.Name)),
170 {std::move(*Edit)}});
171 }
172 } else {
173 vlog("Failed to calculate include insertion for {0} into {1}: {2}", Inc,
174 File, ToInclude.takeError());
175 }
176 }
177 }
178 return Fixes;
179 }
180
181 // Returns the identifiers qualified by an unresolved name. \p Loc is the
182 // start location of the unresolved name. For the example below, this returns
183 // "::X::Y" that is qualified by unresolved name "clangd":
184 // clang::clangd::X::Y
185 // ~
qualifiedByUnresolved(const SourceManager & SM,SourceLocation Loc,const LangOptions & LangOpts)186 llvm::Optional<std::string> qualifiedByUnresolved(const SourceManager &SM,
187 SourceLocation Loc,
188 const LangOptions &LangOpts) {
189 std::string Result;
190
191 SourceLocation NextLoc = Loc;
192 while (auto CCTok = Lexer::findNextToken(NextLoc, SM, LangOpts)) {
193 if (!CCTok->is(tok::coloncolon))
194 break;
195 auto IDTok = Lexer::findNextToken(CCTok->getLocation(), SM, LangOpts);
196 if (!IDTok || !IDTok->is(tok::raw_identifier))
197 break;
198 Result.append(("::" + IDTok->getRawIdentifier()).str());
199 NextLoc = IDTok->getLocation();
200 }
201 if (Result.empty())
202 return llvm::None;
203 return Result;
204 }
205
206 // An unresolved name and its scope information that can be extracted cheaply.
207 struct CheapUnresolvedName {
208 std::string Name;
209 // This is the part of what was typed that was resolved, and it's in its
210 // resolved form not its typed form (think `namespace clang { clangd::x }` -->
211 // `clang::clangd::`).
212 llvm::Optional<std::string> ResolvedScope;
213
214 // Unresolved part of the scope. When the unresolved name is a specifier, we
215 // use the name that comes after it as the alternative name to resolve and use
216 // the specifier as the extra scope in the accessible scopes.
217 llvm::Optional<std::string> UnresolvedScope;
218 };
219
220 // Extracts unresolved name and scope information around \p Unresolved.
221 // FIXME: try to merge this with the scope-wrangling code in CodeComplete.
extractUnresolvedNameCheaply(const SourceManager & SM,const DeclarationNameInfo & Unresolved,CXXScopeSpec * SS,const LangOptions & LangOpts,bool UnresolvedIsSpecifier)222 llvm::Optional<CheapUnresolvedName> extractUnresolvedNameCheaply(
223 const SourceManager &SM, const DeclarationNameInfo &Unresolved,
224 CXXScopeSpec *SS, const LangOptions &LangOpts, bool UnresolvedIsSpecifier) {
225 bool Invalid = false;
226 llvm::StringRef Code = SM.getBufferData(
227 SM.getDecomposedLoc(Unresolved.getBeginLoc()).first, &Invalid);
228 if (Invalid)
229 return llvm::None;
230 CheapUnresolvedName Result;
231 Result.Name = Unresolved.getAsString();
232 if (SS && SS->isNotEmpty()) { // "::" or "ns::"
233 if (auto *Nested = SS->getScopeRep()) {
234 if (Nested->getKind() == NestedNameSpecifier::Global)
235 Result.ResolvedScope = "";
236 else if (const auto *NS = Nested->getAsNamespace()) {
237 auto SpecifiedNS = printNamespaceScope(*NS);
238
239 // Check the specifier spelled in the source.
240 // If the resolved scope doesn't end with the spelled scope. The
241 // resolved scope can come from a sema typo correction. For example,
242 // sema assumes that "clangd::" is a typo of "clang::" and uses
243 // "clang::" as the specified scope in:
244 // namespace clang { clangd::X; }
245 // In this case, we use the "typo" specifier as extra scope instead
246 // of using the scope assumed by sema.
247 auto B = SM.getFileOffset(SS->getBeginLoc());
248 auto E = SM.getFileOffset(SS->getEndLoc());
249 std::string Spelling = (Code.substr(B, E - B) + "::").str();
250 if (llvm::StringRef(SpecifiedNS).endswith(Spelling))
251 Result.ResolvedScope = SpecifiedNS;
252 else
253 Result.UnresolvedScope = Spelling;
254 } else if (const auto *ANS = Nested->getAsNamespaceAlias()) {
255 Result.ResolvedScope = printNamespaceScope(*ANS->getNamespace());
256 } else {
257 // We don't fix symbols in scopes that are not top-level e.g. class
258 // members, as we don't collect includes for them.
259 return llvm::None;
260 }
261 }
262 }
263
264 if (UnresolvedIsSpecifier) {
265 // If the unresolved name is a specifier e.g.
266 // clang::clangd::X
267 // ~~~~~~
268 // We try to resolve clang::clangd::X instead of clang::clangd.
269 // FIXME: We won't be able to fix include if the specifier is what we
270 // should resolve (e.g. it's a class scope specifier). Collecting include
271 // headers for nested types could make this work.
272
273 // Not using the end location as it doesn't always point to the end of
274 // identifier.
275 if (auto QualifiedByUnresolved =
276 qualifiedByUnresolved(SM, Unresolved.getBeginLoc(), LangOpts)) {
277 auto Split = splitQualifiedName(*QualifiedByUnresolved);
278 if (!Result.UnresolvedScope)
279 Result.UnresolvedScope.emplace();
280 // If UnresolvedSpecifiedScope is already set, we simply append the
281 // extra scope. Suppose the unresolved name is "index" in the following
282 // example:
283 // namespace clang { clangd::index::X; }
284 // ~~~~~~ ~~~~~
285 // "clangd::" is assumed to be clang:: by Sema, and we would have used
286 // it as extra scope. With "index" being a specifier, we append "index::"
287 // to the extra scope.
288 Result.UnresolvedScope->append((Result.Name + Split.first).str());
289 Result.Name = std::string(Split.second);
290 }
291 }
292 return Result;
293 }
294
295 /// Returns all namespace scopes that the unqualified lookup would visit.
296 std::vector<std::string>
collectAccessibleScopes(Sema & Sem,const DeclarationNameInfo & Typo,Scope * S,Sema::LookupNameKind LookupKind)297 collectAccessibleScopes(Sema &Sem, const DeclarationNameInfo &Typo, Scope *S,
298 Sema::LookupNameKind LookupKind) {
299 // Collects contexts visited during a Sema name lookup.
300 struct VisitedContextCollector : public VisibleDeclConsumer {
301 VisitedContextCollector(std::vector<std::string> &Out) : Out(Out) {}
302 void EnteredContext(DeclContext *Ctx) override {
303 if (llvm::isa<NamespaceDecl>(Ctx))
304 Out.push_back(printNamespaceScope(*Ctx));
305 }
306 void FoundDecl(NamedDecl *ND, NamedDecl *Hiding, DeclContext *Ctx,
307 bool InBaseClass) override {}
308 std::vector<std::string> &Out;
309 };
310
311 std::vector<std::string> Scopes;
312 Scopes.push_back("");
313 VisitedContextCollector Collector(Scopes);
314 Sem.LookupVisibleDecls(S, LookupKind, Collector,
315 /*IncludeGlobalScope=*/false,
316 /*LoadExternal=*/false);
317 std::sort(Scopes.begin(), Scopes.end());
318 Scopes.erase(std::unique(Scopes.begin(), Scopes.end()), Scopes.end());
319 return Scopes;
320 }
321
322 class IncludeFixer::UnresolvedNameRecorder : public ExternalSemaSource {
323 public:
UnresolvedNameRecorder(llvm::Optional<UnresolvedName> & LastUnresolvedName)324 UnresolvedNameRecorder(llvm::Optional<UnresolvedName> &LastUnresolvedName)
325 : LastUnresolvedName(LastUnresolvedName) {}
326
InitializeSema(Sema & S)327 void InitializeSema(Sema &S) override { this->SemaPtr = &S; }
328
329 // Captures the latest typo and treat it as an unresolved name that can
330 // potentially be fixed by adding #includes.
CorrectTypo(const DeclarationNameInfo & Typo,int LookupKind,Scope * S,CXXScopeSpec * SS,CorrectionCandidateCallback & CCC,DeclContext * MemberContext,bool EnteringContext,const ObjCObjectPointerType * OPT)331 TypoCorrection CorrectTypo(const DeclarationNameInfo &Typo, int LookupKind,
332 Scope *S, CXXScopeSpec *SS,
333 CorrectionCandidateCallback &CCC,
334 DeclContext *MemberContext, bool EnteringContext,
335 const ObjCObjectPointerType *OPT) override {
336 assert(SemaPtr && "Sema must have been set.");
337 if (SemaPtr->isSFINAEContext())
338 return TypoCorrection();
339 if (!isInsideMainFile(Typo.getLoc(), SemaPtr->SourceMgr))
340 return clang::TypoCorrection();
341
342 auto Extracted = extractUnresolvedNameCheaply(
343 SemaPtr->SourceMgr, Typo, SS, SemaPtr->LangOpts,
344 static_cast<Sema::LookupNameKind>(LookupKind) ==
345 Sema::LookupNameKind::LookupNestedNameSpecifierName);
346 if (!Extracted)
347 return TypoCorrection();
348
349 UnresolvedName Unresolved;
350 Unresolved.Name = Extracted->Name;
351 Unresolved.Loc = Typo.getBeginLoc();
352 if (!Extracted->ResolvedScope && !S) // Give up if no scope available.
353 return TypoCorrection();
354
355 if (Extracted->ResolvedScope)
356 Unresolved.Scopes.push_back(*Extracted->ResolvedScope);
357 else // no qualifier or qualifier is unresolved.
358 Unresolved.Scopes = collectAccessibleScopes(
359 *SemaPtr, Typo, S, static_cast<Sema::LookupNameKind>(LookupKind));
360
361 if (Extracted->UnresolvedScope) {
362 for (std::string &Scope : Unresolved.Scopes)
363 Scope += *Extracted->UnresolvedScope;
364 }
365
366 LastUnresolvedName = std::move(Unresolved);
367
368 // Never return a valid correction to try to recover. Our suggested fixes
369 // always require a rebuild.
370 return TypoCorrection();
371 }
372
373 private:
374 Sema *SemaPtr = nullptr;
375
376 llvm::Optional<UnresolvedName> &LastUnresolvedName;
377 };
378
379 llvm::IntrusiveRefCntPtr<ExternalSemaSource>
unresolvedNameRecorder()380 IncludeFixer::unresolvedNameRecorder() {
381 return new UnresolvedNameRecorder(LastUnresolvedName);
382 }
383
fixUnresolvedName() const384 std::vector<Fix> IncludeFixer::fixUnresolvedName() const {
385 assert(LastUnresolvedName.hasValue());
386 auto &Unresolved = *LastUnresolvedName;
387 vlog("Trying to fix unresolved name \"{0}\" in scopes: [{1}]",
388 Unresolved.Name, llvm::join(Unresolved.Scopes, ", "));
389
390 FuzzyFindRequest Req;
391 Req.AnyScope = false;
392 Req.Query = Unresolved.Name;
393 Req.Scopes = Unresolved.Scopes;
394 Req.RestrictForCodeCompletion = true;
395 Req.Limit = 100;
396
397 if (llvm::Optional<const SymbolSlab *> Syms = fuzzyFindCached(Req))
398 return fixesForSymbols(**Syms);
399
400 return {};
401 }
402
403 llvm::Optional<const SymbolSlab *>
fuzzyFindCached(const FuzzyFindRequest & Req) const404 IncludeFixer::fuzzyFindCached(const FuzzyFindRequest &Req) const {
405 auto ReqStr = llvm::formatv("{0}", toJSON(Req)).str();
406 auto I = FuzzyFindCache.find(ReqStr);
407 if (I != FuzzyFindCache.end())
408 return &I->second;
409
410 if (IndexRequestCount >= IndexRequestLimit)
411 return llvm::None;
412 IndexRequestCount++;
413
414 SymbolSlab::Builder Matches;
415 Index.fuzzyFind(Req, [&](const Symbol &Sym) {
416 if (Sym.Name != Req.Query)
417 return;
418 if (!Sym.IncludeHeaders.empty())
419 Matches.insert(Sym);
420 });
421 auto Syms = std::move(Matches).build();
422 auto E = FuzzyFindCache.try_emplace(ReqStr, std::move(Syms));
423 return &E.first->second;
424 }
425
426 llvm::Optional<const SymbolSlab *>
lookupCached(const SymbolID & ID) const427 IncludeFixer::lookupCached(const SymbolID &ID) const {
428 LookupRequest Req;
429 Req.IDs.insert(ID);
430
431 auto I = LookupCache.find(ID);
432 if (I != LookupCache.end())
433 return &I->second;
434
435 if (IndexRequestCount >= IndexRequestLimit)
436 return llvm::None;
437 IndexRequestCount++;
438
439 // FIXME: consider batching the requests for all diagnostics.
440 SymbolSlab::Builder Matches;
441 Index.lookup(Req, [&](const Symbol &Sym) { Matches.insert(Sym); });
442 auto Syms = std::move(Matches).build();
443
444 std::vector<Fix> Fixes;
445 if (!Syms.empty()) {
446 auto &Matched = *Syms.begin();
447 if (!Matched.IncludeHeaders.empty() && Matched.Definition &&
448 Matched.CanonicalDeclaration.FileURI == Matched.Definition.FileURI)
449 Fixes = fixesForSymbols(Syms);
450 }
451 auto E = LookupCache.try_emplace(ID, std::move(Syms));
452 return &E.first->second;
453 }
454
455 } // namespace clangd
456 } // namespace clang
457