xref: /llvm-project/clang/lib/ExtractAPI/Serialization/SymbolGraphSerializer.cpp (revision 80b787e803292119f30da2e1e95acff5beea61db)
1 //===- ExtractAPI/Serialization/SymbolGraphSerializer.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 /// \file
10 /// This file implements the SymbolGraphSerializer.
11 ///
12 //===----------------------------------------------------------------------===//
13 
14 #include "clang/ExtractAPI/Serialization/SymbolGraphSerializer.h"
15 #include "clang/Basic/SourceLocation.h"
16 #include "clang/Basic/Version.h"
17 #include "clang/ExtractAPI/DeclarationFragments.h"
18 #include "llvm/ADT/STLExtras.h"
19 #include "llvm/ADT/STLFunctionalExtras.h"
20 #include "llvm/Support/Casting.h"
21 #include "llvm/Support/Compiler.h"
22 #include "llvm/Support/Path.h"
23 #include "llvm/Support/VersionTuple.h"
24 #include <optional>
25 #include <type_traits>
26 
27 using namespace clang;
28 using namespace clang::extractapi;
29 using namespace llvm;
30 using namespace llvm::json;
31 
32 namespace {
33 
34 /// Helper function to inject a JSON object \p Obj into another object \p Paren
35 /// at position \p Key.
36 void serializeObject(Object &Paren, StringRef Key, std::optional<Object> Obj) {
37   if (Obj)
38     Paren[Key] = std::move(*Obj);
39 }
40 
41 /// Helper function to inject a StringRef \p String into an object \p Paren at
42 /// position \p Key
43 void serializeString(Object &Paren, StringRef Key,
44                      std::optional<std::string> String) {
45   if (String)
46     Paren[Key] = std::move(*String);
47 }
48 
49 /// Helper function to inject a JSON array \p Array into object \p Paren at
50 /// position \p Key.
51 void serializeArray(Object &Paren, StringRef Key, std::optional<Array> Array) {
52   if (Array)
53     Paren[Key] = std::move(*Array);
54 }
55 
56 /// Serialize a \c VersionTuple \p V with the Symbol Graph semantic version
57 /// format.
58 ///
59 /// A semantic version object contains three numeric fields, representing the
60 /// \c major, \c minor, and \c patch parts of the version tuple.
61 /// For example version tuple 1.0.3 is serialized as:
62 /// \code
63 ///   {
64 ///     "major" : 1,
65 ///     "minor" : 0,
66 ///     "patch" : 3
67 ///   }
68 /// \endcode
69 ///
70 /// \returns \c std::nullopt if the version \p V is empty, or an \c Object
71 /// containing the semantic version representation of \p V.
72 std::optional<Object> serializeSemanticVersion(const VersionTuple &V) {
73   if (V.empty())
74     return std::nullopt;
75 
76   Object Version;
77   Version["major"] = V.getMajor();
78   Version["minor"] = V.getMinor().value_or(0);
79   Version["patch"] = V.getSubminor().value_or(0);
80   return Version;
81 }
82 
83 /// Serialize the OS information in the Symbol Graph platform property.
84 ///
85 /// The OS information in Symbol Graph contains the \c name of the OS, and an
86 /// optional \c minimumVersion semantic version field.
87 Object serializeOperatingSystem(const Triple &T) {
88   Object OS;
89   OS["name"] = T.getOSTypeName(T.getOS());
90   serializeObject(OS, "minimumVersion",
91                   serializeSemanticVersion(T.getMinimumSupportedOSVersion()));
92   return OS;
93 }
94 
95 /// Serialize the platform information in the Symbol Graph module section.
96 ///
97 /// The platform object describes a target platform triple in corresponding
98 /// three fields: \c architecture, \c vendor, and \c operatingSystem.
99 Object serializePlatform(const Triple &T) {
100   Object Platform;
101   Platform["architecture"] = T.getArchName();
102   Platform["vendor"] = T.getVendorName();
103   Platform["operatingSystem"] = serializeOperatingSystem(T);
104   return Platform;
105 }
106 
107 /// Serialize a source position.
108 Object serializeSourcePosition(const PresumedLoc &Loc) {
109   assert(Loc.isValid() && "invalid source position");
110 
111   Object SourcePosition;
112   SourcePosition["line"] = Loc.getLine();
113   SourcePosition["character"] = Loc.getColumn();
114 
115   return SourcePosition;
116 }
117 
118 /// Serialize a source location in file.
119 ///
120 /// \param Loc The presumed location to serialize.
121 /// \param IncludeFileURI If true, include the file path of \p Loc as a URI.
122 /// Defaults to false.
123 Object serializeSourceLocation(const PresumedLoc &Loc,
124                                bool IncludeFileURI = false) {
125   Object SourceLocation;
126   serializeObject(SourceLocation, "position", serializeSourcePosition(Loc));
127 
128   if (IncludeFileURI) {
129     std::string FileURI = "file://";
130     // Normalize file path to use forward slashes for the URI.
131     FileURI += sys::path::convert_to_slash(Loc.getFilename());
132     SourceLocation["uri"] = FileURI;
133   }
134 
135   return SourceLocation;
136 }
137 
138 /// Serialize a source range with begin and end locations.
139 Object serializeSourceRange(const PresumedLoc &BeginLoc,
140                             const PresumedLoc &EndLoc) {
141   Object SourceRange;
142   serializeObject(SourceRange, "start", serializeSourcePosition(BeginLoc));
143   serializeObject(SourceRange, "end", serializeSourcePosition(EndLoc));
144   return SourceRange;
145 }
146 
147 /// Serialize the availability attributes of a symbol.
148 ///
149 /// Availability information contains the introduced, deprecated, and obsoleted
150 /// versions of the symbol for a given domain (roughly corresponds to a
151 /// platform) as semantic versions, if not default.  Availability information
152 /// also contains flags to indicate if the symbol is unconditionally unavailable
153 /// or deprecated, i.e. \c __attribute__((unavailable)) and \c
154 /// __attribute__((deprecated)).
155 ///
156 /// \returns \c std::nullopt if the symbol has default availability attributes,
157 /// or an \c Array containing the formatted availability information.
158 std::optional<Array>
159 serializeAvailability(const AvailabilitySet &Availabilities) {
160   if (Availabilities.isDefault())
161     return std::nullopt;
162 
163   Array AvailabilityArray;
164 
165   if (Availabilities.isUnconditionallyDeprecated()) {
166     Object UnconditionallyDeprecated;
167     UnconditionallyDeprecated["domain"] = "*";
168     UnconditionallyDeprecated["isUnconditionallyDeprecated"] = true;
169     AvailabilityArray.emplace_back(std::move(UnconditionallyDeprecated));
170   }
171 
172   // Note unconditionally unavailable records are skipped.
173 
174   for (const auto &AvailInfo : Availabilities) {
175     Object Availability;
176     Availability["domain"] = AvailInfo.Domain;
177     if (AvailInfo.Unavailable)
178       Availability["isUnconditionallyUnavailable"] = true;
179     else {
180       serializeObject(Availability, "introducedVersion",
181                       serializeSemanticVersion(AvailInfo.Introduced));
182       serializeObject(Availability, "deprecatedVersion",
183                       serializeSemanticVersion(AvailInfo.Deprecated));
184       serializeObject(Availability, "obsoletedVersion",
185                       serializeSemanticVersion(AvailInfo.Obsoleted));
186     }
187     AvailabilityArray.emplace_back(std::move(Availability));
188   }
189 
190   return AvailabilityArray;
191 }
192 
193 /// Get the language name string for interface language references.
194 StringRef getLanguageName(Language Lang) {
195   switch (Lang) {
196   case Language::C:
197     return "c";
198   case Language::ObjC:
199     return "objective-c";
200   case Language::CXX:
201     return "c++";
202 
203   // Unsupported language currently
204   case Language::ObjCXX:
205   case Language::OpenCL:
206   case Language::OpenCLCXX:
207   case Language::CUDA:
208   case Language::RenderScript:
209   case Language::HIP:
210   case Language::HLSL:
211 
212   // Languages that the frontend cannot parse and compile
213   case Language::Unknown:
214   case Language::Asm:
215   case Language::LLVM_IR:
216     llvm_unreachable("Unsupported language kind");
217   }
218 
219   llvm_unreachable("Unhandled language kind");
220 }
221 
222 /// Serialize the identifier object as specified by the Symbol Graph format.
223 ///
224 /// The identifier property of a symbol contains the USR for precise and unique
225 /// references, and the interface language name.
226 Object serializeIdentifier(const APIRecord &Record, Language Lang) {
227   Object Identifier;
228   Identifier["precise"] = Record.USR;
229   Identifier["interfaceLanguage"] = getLanguageName(Lang);
230 
231   return Identifier;
232 }
233 
234 /// Serialize the documentation comments attached to a symbol, as specified by
235 /// the Symbol Graph format.
236 ///
237 /// The Symbol Graph \c docComment object contains an array of lines. Each line
238 /// represents one line of striped documentation comment, with source range
239 /// information.
240 /// e.g.
241 /// \code
242 ///   /// This is a documentation comment
243 ///       ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~'  First line.
244 ///   ///     with multiple lines.
245 ///       ^~~~~~~~~~~~~~~~~~~~~~~'         Second line.
246 /// \endcode
247 ///
248 /// \returns \c std::nullopt if \p Comment is empty, or an \c Object containing
249 /// the formatted lines.
250 std::optional<Object> serializeDocComment(const DocComment &Comment) {
251   if (Comment.empty())
252     return std::nullopt;
253 
254   Object DocComment;
255   Array LinesArray;
256   for (const auto &CommentLine : Comment) {
257     Object Line;
258     Line["text"] = CommentLine.Text;
259     serializeObject(Line, "range",
260                     serializeSourceRange(CommentLine.Begin, CommentLine.End));
261     LinesArray.emplace_back(std::move(Line));
262   }
263   serializeArray(DocComment, "lines", LinesArray);
264 
265   return DocComment;
266 }
267 
268 /// Serialize the declaration fragments of a symbol.
269 ///
270 /// The Symbol Graph declaration fragments is an array of tagged important
271 /// parts of a symbol's declaration. The fragments sequence can be joined to
272 /// form spans of declaration text, with attached information useful for
273 /// purposes like syntax-highlighting etc. For example:
274 /// \code
275 ///   const int pi; -> "declarationFragments" : [
276 ///                      {
277 ///                        "kind" : "keyword",
278 ///                        "spelling" : "const"
279 ///                      },
280 ///                      {
281 ///                        "kind" : "text",
282 ///                        "spelling" : " "
283 ///                      },
284 ///                      {
285 ///                        "kind" : "typeIdentifier",
286 ///                        "preciseIdentifier" : "c:I",
287 ///                        "spelling" : "int"
288 ///                      },
289 ///                      {
290 ///                        "kind" : "text",
291 ///                        "spelling" : " "
292 ///                      },
293 ///                      {
294 ///                        "kind" : "identifier",
295 ///                        "spelling" : "pi"
296 ///                      }
297 ///                    ]
298 /// \endcode
299 ///
300 /// \returns \c std::nullopt if \p DF is empty, or an \c Array containing the
301 /// formatted declaration fragments array.
302 std::optional<Array>
303 serializeDeclarationFragments(const DeclarationFragments &DF) {
304   if (DF.getFragments().empty())
305     return std::nullopt;
306 
307   Array Fragments;
308   for (const auto &F : DF.getFragments()) {
309     Object Fragment;
310     Fragment["spelling"] = F.Spelling;
311     Fragment["kind"] = DeclarationFragments::getFragmentKindString(F.Kind);
312     if (!F.PreciseIdentifier.empty())
313       Fragment["preciseIdentifier"] = F.PreciseIdentifier;
314     Fragments.emplace_back(std::move(Fragment));
315   }
316 
317   return Fragments;
318 }
319 
320 /// Serialize the \c names field of a symbol as specified by the Symbol Graph
321 /// format.
322 ///
323 /// The Symbol Graph names field contains multiple representations of a symbol
324 /// that can be used for different applications:
325 ///   - \c title : The simple declared name of the symbol;
326 ///   - \c subHeading : An array of declaration fragments that provides tags,
327 ///     and potentially more tokens (for example the \c +/- symbol for
328 ///     Objective-C methods). Can be used as sub-headings for documentation.
329 Object serializeNames(const APIRecord &Record) {
330   Object Names;
331   if (auto *CategoryRecord =
332           dyn_cast_or_null<const ObjCCategoryRecord>(&Record))
333     Names["title"] =
334         (CategoryRecord->Interface.Name + " (" + Record.Name + ")").str();
335   else
336     Names["title"] = Record.Name;
337 
338   serializeArray(Names, "subHeading",
339                  serializeDeclarationFragments(Record.SubHeading));
340   DeclarationFragments NavigatorFragments;
341   NavigatorFragments.append(Record.Name,
342                             DeclarationFragments::FragmentKind::Identifier,
343                             /*PreciseIdentifier*/ "");
344   serializeArray(Names, "navigator",
345                  serializeDeclarationFragments(NavigatorFragments));
346 
347   return Names;
348 }
349 
350 Object serializeSymbolKind(APIRecord::RecordKind RK, Language Lang) {
351   auto AddLangPrefix = [&Lang](StringRef S) -> std::string {
352     return (getLanguageName(Lang) + "." + S).str();
353   };
354 
355   Object Kind;
356   switch (RK) {
357   case APIRecord::RK_Unknown:
358     llvm_unreachable("Records should have an explicit kind");
359     break;
360   case APIRecord::RK_GlobalFunction:
361     Kind["identifier"] = AddLangPrefix("func");
362     Kind["displayName"] = "Function";
363     break;
364   case APIRecord::RK_GlobalFunctionTemplate:
365     Kind["identifier"] = AddLangPrefix("func");
366     Kind["displayName"] = "Function Template";
367     break;
368   case APIRecord::RK_GlobalFunctionTemplateSpecialization:
369     Kind["identifier"] = AddLangPrefix("func");
370     Kind["displayName"] = "Function Template Specialization";
371     break;
372   case APIRecord::RK_GlobalVariableTemplate:
373     Kind["identifier"] = AddLangPrefix("var");
374     Kind["displayName"] = "Global Variable Template";
375     break;
376   case APIRecord::RK_GlobalVariableTemplateSpecialization:
377     Kind["identifier"] = AddLangPrefix("var");
378     Kind["displayName"] = "Global Variable Template Specialization";
379     break;
380   case APIRecord::RK_GlobalVariableTemplatePartialSpecialization:
381     Kind["identifier"] = AddLangPrefix("var");
382     Kind["displayName"] = "Global Variable Template Partial Specialization";
383     break;
384   case APIRecord::RK_GlobalVariable:
385     Kind["identifier"] = AddLangPrefix("var");
386     Kind["displayName"] = "Global Variable";
387     break;
388   case APIRecord::RK_EnumConstant:
389     Kind["identifier"] = AddLangPrefix("enum.case");
390     Kind["displayName"] = "Enumeration Case";
391     break;
392   case APIRecord::RK_Enum:
393     Kind["identifier"] = AddLangPrefix("enum");
394     Kind["displayName"] = "Enumeration";
395     break;
396   case APIRecord::RK_StructField:
397     Kind["identifier"] = AddLangPrefix("property");
398     Kind["displayName"] = "Instance Property";
399     break;
400   case APIRecord::RK_Struct:
401     Kind["identifier"] = AddLangPrefix("struct");
402     Kind["displayName"] = "Structure";
403     break;
404   case APIRecord::RK_CXXField:
405     Kind["identifier"] = AddLangPrefix("property");
406     Kind["displayName"] = "Instance Property";
407     break;
408   case APIRecord::RK_Union:
409     Kind["identifier"] = AddLangPrefix("union");
410     Kind["displayName"] = "Union";
411     break;
412   case APIRecord::RK_StaticField:
413     Kind["identifier"] = AddLangPrefix("type.property");
414     Kind["displayName"] = "Type Property";
415     break;
416   case APIRecord::RK_ClassTemplate:
417   case APIRecord::RK_ClassTemplateSpecialization:
418   case APIRecord::RK_ClassTemplatePartialSpecialization:
419   case APIRecord::RK_CXXClass:
420     Kind["identifier"] = AddLangPrefix("class");
421     Kind["displayName"] = "Class";
422     break;
423   case APIRecord::RK_Concept:
424     Kind["identifier"] = AddLangPrefix("concept");
425     Kind["displayName"] = "Concept";
426     break;
427   case APIRecord::RK_CXXStaticMethod:
428     Kind["identifier"] = AddLangPrefix("type.method");
429     Kind["displayName"] = "Static Method";
430     break;
431   case APIRecord::RK_CXXInstanceMethod:
432     Kind["identifier"] = AddLangPrefix("method");
433     Kind["displayName"] = "Instance Method";
434     break;
435   case APIRecord::RK_CXXConstructorMethod:
436     Kind["identifier"] = AddLangPrefix("method");
437     Kind["displayName"] = "Constructor";
438     break;
439   case APIRecord::RK_CXXDestructorMethod:
440     Kind["identifier"] = AddLangPrefix("method");
441     Kind["displayName"] = "Destructor";
442     break;
443   case APIRecord::RK_ObjCIvar:
444     Kind["identifier"] = AddLangPrefix("ivar");
445     Kind["displayName"] = "Instance Variable";
446     break;
447   case APIRecord::RK_ObjCInstanceMethod:
448     Kind["identifier"] = AddLangPrefix("method");
449     Kind["displayName"] = "Instance Method";
450     break;
451   case APIRecord::RK_ObjCClassMethod:
452     Kind["identifier"] = AddLangPrefix("type.method");
453     Kind["displayName"] = "Type Method";
454     break;
455   case APIRecord::RK_ObjCInstanceProperty:
456     Kind["identifier"] = AddLangPrefix("property");
457     Kind["displayName"] = "Instance Property";
458     break;
459   case APIRecord::RK_ObjCClassProperty:
460     Kind["identifier"] = AddLangPrefix("type.property");
461     Kind["displayName"] = "Type Property";
462     break;
463   case APIRecord::RK_ObjCInterface:
464     Kind["identifier"] = AddLangPrefix("class");
465     Kind["displayName"] = "Class";
466     break;
467   case APIRecord::RK_ObjCCategory:
468     Kind["identifier"] = AddLangPrefix("class.extension");
469     Kind["displayName"] = "Class Extension";
470     break;
471   case APIRecord::RK_ObjCCategoryModule:
472     Kind["identifier"] = AddLangPrefix("module.extension");
473     Kind["displayName"] = "Module Extension";
474     break;
475   case APIRecord::RK_ObjCProtocol:
476     Kind["identifier"] = AddLangPrefix("protocol");
477     Kind["displayName"] = "Protocol";
478     break;
479   case APIRecord::RK_MacroDefinition:
480     Kind["identifier"] = AddLangPrefix("macro");
481     Kind["displayName"] = "Macro";
482     break;
483   case APIRecord::RK_Typedef:
484     Kind["identifier"] = AddLangPrefix("typealias");
485     Kind["displayName"] = "Type Alias";
486     break;
487   }
488 
489   return Kind;
490 }
491 
492 /// Serialize the symbol kind information.
493 ///
494 /// The Symbol Graph symbol kind property contains a shorthand \c identifier
495 /// which is prefixed by the source language name, useful for tooling to parse
496 /// the kind, and a \c displayName for rendering human-readable names.
497 Object serializeSymbolKind(const APIRecord &Record, Language Lang) {
498   return serializeSymbolKind(Record.getKind(), Lang);
499 }
500 
501 template <typename RecordTy>
502 std::optional<Object>
503 serializeFunctionSignatureMixinImpl(const RecordTy &Record, std::true_type) {
504   const auto &FS = Record.Signature;
505   if (FS.empty())
506     return std::nullopt;
507 
508   Object Signature;
509   serializeArray(Signature, "returns",
510                  serializeDeclarationFragments(FS.getReturnType()));
511 
512   Array Parameters;
513   for (const auto &P : FS.getParameters()) {
514     Object Parameter;
515     Parameter["name"] = P.Name;
516     serializeArray(Parameter, "declarationFragments",
517                    serializeDeclarationFragments(P.Fragments));
518     Parameters.emplace_back(std::move(Parameter));
519   }
520 
521   if (!Parameters.empty())
522     Signature["parameters"] = std::move(Parameters);
523 
524   return Signature;
525 }
526 
527 template <typename RecordTy>
528 std::optional<Object>
529 serializeFunctionSignatureMixinImpl(const RecordTy &Record, std::false_type) {
530   return std::nullopt;
531 }
532 
533 /// Serialize the function signature field, as specified by the
534 /// Symbol Graph format.
535 ///
536 /// The Symbol Graph function signature property contains two arrays.
537 ///   - The \c returns array is the declaration fragments of the return type;
538 ///   - The \c parameters array contains names and declaration fragments of the
539 ///     parameters.
540 ///
541 /// \returns \c std::nullopt if \p FS is empty, or an \c Object containing the
542 /// formatted function signature.
543 template <typename RecordTy>
544 void serializeFunctionSignatureMixin(Object &Paren, const RecordTy &Record) {
545   serializeObject(Paren, "functionSignature",
546                   serializeFunctionSignatureMixinImpl(
547                       Record, has_function_signature<RecordTy>()));
548 }
549 
550 template <typename RecordTy>
551 std::optional<std::string> serializeAccessMixinImpl(const RecordTy &Record,
552                                                     std::true_type) {
553   const auto &AccessControl = Record.Access;
554   std::string Access;
555   if (AccessControl.empty())
556     return std::nullopt;
557   Access = AccessControl.getAccess();
558   return Access;
559 }
560 
561 template <typename RecordTy>
562 std::optional<std::string> serializeAccessMixinImpl(const RecordTy &Record,
563                                                     std::false_type) {
564   return std::nullopt;
565 }
566 
567 template <typename RecordTy>
568 void serializeAccessMixin(Object &Paren, const RecordTy &Record) {
569   auto accessLevel = serializeAccessMixinImpl(Record, has_access<RecordTy>());
570   if (!accessLevel.has_value())
571     accessLevel = "public";
572   serializeString(Paren, "accessLevel", accessLevel);
573 }
574 
575 template <typename RecordTy>
576 std::optional<Object> serializeTemplateMixinImpl(const RecordTy &Record,
577                                                  std::true_type) {
578   const auto &Template = Record.Templ;
579   if (Template.empty())
580     return std::nullopt;
581 
582   Object Generics;
583   Array GenericParameters;
584   for (const auto Param : Template.getParameters()) {
585     Object Parameter;
586     Parameter["name"] = Param.Name;
587     Parameter["index"] = Param.Index;
588     Parameter["depth"] = Param.Depth;
589     GenericParameters.emplace_back(std::move(Parameter));
590   }
591   if (!GenericParameters.empty())
592     Generics["parameters"] = std::move(GenericParameters);
593 
594   Array GenericConstraints;
595   for (const auto Constr : Template.getConstraints()) {
596     Object Constraint;
597     Constraint["kind"] = Constr.Kind;
598     Constraint["lhs"] = Constr.LHS;
599     Constraint["rhs"] = Constr.RHS;
600     GenericConstraints.emplace_back(std::move(Constraint));
601   }
602 
603   if (!GenericConstraints.empty())
604     Generics["constraints"] = std::move(GenericConstraints);
605 
606   return Generics;
607 }
608 
609 template <typename RecordTy>
610 std::optional<Object> serializeTemplateMixinImpl(const RecordTy &Record,
611                                                  std::false_type) {
612   return std::nullopt;
613 }
614 
615 template <typename RecordTy>
616 void serializeTemplateMixin(Object &Paren, const RecordTy &Record) {
617   serializeObject(Paren, "swiftGenerics",
618                   serializeTemplateMixinImpl(Record, has_template<RecordTy>()));
619 }
620 
621 struct PathComponent {
622   StringRef USR;
623   StringRef Name;
624   APIRecord::RecordKind Kind;
625 
626   PathComponent(StringRef USR, StringRef Name, APIRecord::RecordKind Kind)
627       : USR(USR), Name(Name), Kind(Kind) {}
628 };
629 
630 template <typename RecordTy>
631 bool generatePathComponents(
632     const RecordTy &Record, const APISet &API,
633     function_ref<void(const PathComponent &)> ComponentTransformer) {
634   SmallVector<PathComponent, 4> ReverseComponenents;
635   ReverseComponenents.emplace_back(Record.USR, Record.Name, Record.getKind());
636   const auto *CurrentParent = &Record.ParentInformation;
637   bool FailedToFindParent = false;
638   while (CurrentParent && !CurrentParent->empty()) {
639     PathComponent CurrentParentComponent(CurrentParent->ParentUSR,
640                                          CurrentParent->ParentName,
641                                          CurrentParent->ParentKind);
642 
643     auto *ParentRecord = CurrentParent->ParentRecord;
644     // Slow path if we don't have a direct reference to the ParentRecord
645     if (!ParentRecord)
646       ParentRecord = API.findRecordForUSR(CurrentParent->ParentUSR);
647 
648     // If the parent is a category extended from internal module then we need to
649     // pretend this belongs to the associated interface.
650     if (auto *CategoryRecord =
651             dyn_cast_or_null<ObjCCategoryRecord>(ParentRecord)) {
652       if (!CategoryRecord->IsFromExternalModule) {
653         ParentRecord = API.findRecordForUSR(CategoryRecord->Interface.USR);
654         CurrentParentComponent = PathComponent(CategoryRecord->Interface.USR,
655                                                CategoryRecord->Interface.Name,
656                                                APIRecord::RK_ObjCInterface);
657       }
658     }
659 
660     // The parent record doesn't exist which means the symbol shouldn't be
661     // treated as part of the current product.
662     if (!ParentRecord) {
663       FailedToFindParent = true;
664       break;
665     }
666 
667     ReverseComponenents.push_back(std::move(CurrentParentComponent));
668     CurrentParent = &ParentRecord->ParentInformation;
669   }
670 
671   for (const auto &PC : reverse(ReverseComponenents))
672     ComponentTransformer(PC);
673 
674   return FailedToFindParent;
675 }
676 
677 Object serializeParentContext(const PathComponent &PC, Language Lang) {
678   Object ParentContextElem;
679   ParentContextElem["usr"] = PC.USR;
680   ParentContextElem["name"] = PC.Name;
681   ParentContextElem["kind"] = serializeSymbolKind(PC.Kind, Lang)["identifier"];
682   return ParentContextElem;
683 }
684 
685 template <typename RecordTy>
686 Array generateParentContexts(const RecordTy &Record, const APISet &API,
687                              Language Lang) {
688   Array ParentContexts;
689   generatePathComponents(
690       Record, API, [Lang, &ParentContexts](const PathComponent &PC) {
691         ParentContexts.push_back(serializeParentContext(PC, Lang));
692       });
693 
694   return ParentContexts;
695 }
696 } // namespace
697 
698 /// Defines the format version emitted by SymbolGraphSerializer.
699 const VersionTuple SymbolGraphSerializer::FormatVersion{0, 5, 3};
700 
701 Object SymbolGraphSerializer::serializeMetadata() const {
702   Object Metadata;
703   serializeObject(Metadata, "formatVersion",
704                   serializeSemanticVersion(FormatVersion));
705   Metadata["generator"] = clang::getClangFullVersion();
706   return Metadata;
707 }
708 
709 Object SymbolGraphSerializer::serializeModule() const {
710   Object Module;
711   // The user is expected to always pass `--product-name=` on the command line
712   // to populate this field.
713   Module["name"] = API.ProductName;
714   serializeObject(Module, "platform", serializePlatform(API.getTarget()));
715   return Module;
716 }
717 
718 bool SymbolGraphSerializer::shouldSkip(const APIRecord &Record) const {
719   // Skip explicitly ignored symbols.
720   if (IgnoresList.shouldIgnore(Record.Name))
721     return true;
722 
723   // Skip unconditionally unavailable symbols
724   if (Record.Availabilities.isUnconditionallyUnavailable())
725     return true;
726 
727   // Filter out symbols prefixed with an underscored as they are understood to
728   // be symbols clients should not use.
729   if (Record.Name.startswith("_"))
730     return true;
731 
732   return false;
733 }
734 
735 template <typename RecordTy>
736 std::optional<Object>
737 SymbolGraphSerializer::serializeAPIRecord(const RecordTy &Record) const {
738   if (shouldSkip(Record))
739     return std::nullopt;
740 
741   Object Obj;
742   serializeObject(Obj, "identifier",
743                   serializeIdentifier(Record, API.getLanguage()));
744   serializeObject(Obj, "kind", serializeSymbolKind(Record, API.getLanguage()));
745   serializeObject(Obj, "names", serializeNames(Record));
746   serializeObject(
747       Obj, "location",
748       serializeSourceLocation(Record.Location, /*IncludeFileURI=*/true));
749   serializeArray(Obj, "availability",
750                  serializeAvailability(Record.Availabilities));
751   serializeObject(Obj, "docComment", serializeDocComment(Record.Comment));
752   serializeArray(Obj, "declarationFragments",
753                  serializeDeclarationFragments(Record.Declaration));
754   SmallVector<StringRef, 4> PathComponentsNames;
755   // If this returns true it indicates that we couldn't find a symbol in the
756   // hierarchy.
757   if (generatePathComponents(Record, API,
758                              [&PathComponentsNames](const PathComponent &PC) {
759                                PathComponentsNames.push_back(PC.Name);
760                              }))
761     return {};
762 
763   serializeArray(Obj, "pathComponents", Array(PathComponentsNames));
764 
765   serializeFunctionSignatureMixin(Obj, Record);
766   serializeAccessMixin(Obj, Record);
767   serializeTemplateMixin(Obj, Record);
768 
769   return Obj;
770 }
771 
772 template <typename MemberTy>
773 void SymbolGraphSerializer::serializeMembers(
774     const APIRecord &Record,
775     const SmallVector<std::unique_ptr<MemberTy>> &Members) {
776   // Members should not be serialized if we aren't recursing.
777   if (!ShouldRecurse)
778     return;
779   for (const auto &Member : Members) {
780     auto MemberRecord = serializeAPIRecord(*Member);
781     if (!MemberRecord)
782       continue;
783 
784     Symbols.emplace_back(std::move(*MemberRecord));
785     serializeRelationship(RelationshipKind::MemberOf, *Member, Record);
786   }
787 }
788 
789 StringRef SymbolGraphSerializer::getRelationshipString(RelationshipKind Kind) {
790   switch (Kind) {
791   case RelationshipKind::MemberOf:
792     return "memberOf";
793   case RelationshipKind::InheritsFrom:
794     return "inheritsFrom";
795   case RelationshipKind::ConformsTo:
796     return "conformsTo";
797   case RelationshipKind::ExtensionTo:
798     return "extensionTo";
799   }
800   llvm_unreachable("Unhandled relationship kind");
801 }
802 
803 StringRef SymbolGraphSerializer::getConstraintString(ConstraintKind Kind) {
804   switch (Kind) {
805   case ConstraintKind::Conformance:
806     return "conformance";
807   case ConstraintKind::ConditionalConformance:
808     return "conditionalConformance";
809   }
810   llvm_unreachable("Unhandled constraint kind");
811 }
812 
813 void SymbolGraphSerializer::serializeRelationship(RelationshipKind Kind,
814                                                   SymbolReference Source,
815                                                   SymbolReference Target) {
816   Object Relationship;
817   Relationship["source"] = Source.USR;
818   Relationship["target"] = Target.USR;
819   Relationship["targetFallback"] = Target.Name;
820   Relationship["kind"] = getRelationshipString(Kind);
821 
822   Relationships.emplace_back(std::move(Relationship));
823 }
824 
825 void SymbolGraphSerializer::visitGlobalFunctionRecord(
826     const GlobalFunctionRecord &Record) {
827   auto Obj = serializeAPIRecord(Record);
828   if (!Obj)
829     return;
830 
831   Symbols.emplace_back(std::move(*Obj));
832 }
833 
834 void SymbolGraphSerializer::visitGlobalVariableRecord(
835     const GlobalVariableRecord &Record) {
836   auto Obj = serializeAPIRecord(Record);
837   if (!Obj)
838     return;
839 
840   Symbols.emplace_back(std::move(*Obj));
841 }
842 
843 void SymbolGraphSerializer::visitEnumRecord(const EnumRecord &Record) {
844   auto Enum = serializeAPIRecord(Record);
845   if (!Enum)
846     return;
847 
848   Symbols.emplace_back(std::move(*Enum));
849   serializeMembers(Record, Record.Constants);
850 }
851 
852 void SymbolGraphSerializer::visitStructRecord(const StructRecord &Record) {
853   auto Struct = serializeAPIRecord(Record);
854   if (!Struct)
855     return;
856 
857   Symbols.emplace_back(std::move(*Struct));
858   serializeMembers(Record, Record.Fields);
859 }
860 
861 void SymbolGraphSerializer::visitStaticFieldRecord(
862     const StaticFieldRecord &Record) {
863   auto StaticField = serializeAPIRecord(Record);
864   if (!StaticField)
865     return;
866   Symbols.emplace_back(std::move(*StaticField));
867   serializeRelationship(RelationshipKind::MemberOf, Record, Record.Context);
868 }
869 
870 void SymbolGraphSerializer::visitCXXClassRecord(const CXXClassRecord &Record) {
871   auto Class = serializeAPIRecord(Record);
872   if (!Class)
873     return;
874 
875   Symbols.emplace_back(std::move(*Class));
876   serializeMembers(Record, Record.Fields);
877   serializeMembers(Record, Record.Methods);
878 
879   for (const auto Base : Record.Bases)
880     serializeRelationship(RelationshipKind::InheritsFrom, Record, Base);
881 }
882 
883 void SymbolGraphSerializer::visitClassTemplateRecord(
884     const ClassTemplateRecord &Record) {
885   auto Class = serializeAPIRecord(Record);
886   if (!Class)
887     return;
888 
889   Symbols.emplace_back(std::move(*Class));
890   serializeMembers(Record, Record.Fields);
891   serializeMembers(Record, Record.Methods);
892 
893   for (const auto Base : Record.Bases)
894     serializeRelationship(RelationshipKind::InheritsFrom, Record, Base);
895 }
896 
897 void SymbolGraphSerializer::visitClassTemplateSpecializationRecord(
898     const ClassTemplateSpecializationRecord &Record) {
899   auto Class = serializeAPIRecord(Record);
900   if (!Class)
901     return;
902 
903   Symbols.emplace_back(std::move(*Class));
904   serializeMembers(Record, Record.Fields);
905   serializeMembers(Record, Record.Methods);
906 
907   for (const auto Base : Record.Bases)
908     serializeRelationship(RelationshipKind::InheritsFrom, Record, Base);
909 }
910 
911 void SymbolGraphSerializer::visitClassTemplatePartialSpecializationRecord(
912     const ClassTemplatePartialSpecializationRecord &Record) {
913   auto Class = serializeAPIRecord(Record);
914   if (!Class)
915     return;
916 
917   Symbols.emplace_back(std::move(*Class));
918   serializeMembers(Record, Record.Fields);
919   serializeMembers(Record, Record.Methods);
920 
921   for (const auto Base : Record.Bases)
922     serializeRelationship(RelationshipKind::InheritsFrom, Record, Base);
923 }
924 
925 void SymbolGraphSerializer::visitConceptRecord(const ConceptRecord &Record) {
926   auto Concept = serializeAPIRecord(Record);
927   if (!Concept)
928     return;
929 
930   Symbols.emplace_back(std::move(*Concept));
931 }
932 
933 void SymbolGraphSerializer::visitGlobalVariableTemplateRecord(
934     const GlobalVariableTemplateRecord &Record) {
935   auto GlobalVariableTemplate = serializeAPIRecord(Record);
936   if (!GlobalVariableTemplate)
937     return;
938   Symbols.emplace_back(std::move(*GlobalVariableTemplate));
939 }
940 
941 void SymbolGraphSerializer::visitGlobalVariableTemplateSpecializationRecord(
942     const GlobalVariableTemplateSpecializationRecord &Record) {
943   auto GlobalVariableTemplateSpecialization = serializeAPIRecord(Record);
944   if (!GlobalVariableTemplateSpecialization)
945     return;
946   Symbols.emplace_back(std::move(*GlobalVariableTemplateSpecialization));
947 }
948 
949 void SymbolGraphSerializer::
950     visitGlobalVariableTemplatePartialSpecializationRecord(
951         const GlobalVariableTemplatePartialSpecializationRecord &Record) {
952   auto GlobalVariableTemplatePartialSpecialization = serializeAPIRecord(Record);
953   if (!GlobalVariableTemplatePartialSpecialization)
954     return;
955   Symbols.emplace_back(std::move(*GlobalVariableTemplatePartialSpecialization));
956 }
957 
958 void SymbolGraphSerializer::visitGlobalFunctionTemplateRecord(
959     const GlobalFunctionTemplateRecord &Record) {
960   auto GlobalFunctionTemplate = serializeAPIRecord(Record);
961   if (!GlobalFunctionTemplate)
962     return;
963   Symbols.emplace_back(std::move(*GlobalFunctionTemplate));
964 }
965 
966 void SymbolGraphSerializer::visitGlobalFunctionTemplateSpecializationRecord(
967     const GlobalFunctionTemplateSpecializationRecord &Record) {
968   auto GlobalFunctionTemplateSpecialization = serializeAPIRecord(Record);
969   if (!GlobalFunctionTemplateSpecialization)
970     return;
971   Symbols.emplace_back(std::move(*GlobalFunctionTemplateSpecialization));
972 }
973 
974 void SymbolGraphSerializer::visitObjCContainerRecord(
975     const ObjCContainerRecord &Record) {
976   auto ObjCContainer = serializeAPIRecord(Record);
977   if (!ObjCContainer)
978     return;
979 
980   Symbols.emplace_back(std::move(*ObjCContainer));
981 
982   serializeMembers(Record, Record.Ivars);
983   serializeMembers(Record, Record.Methods);
984   serializeMembers(Record, Record.Properties);
985 
986   for (const auto &Protocol : Record.Protocols)
987     // Record that Record conforms to Protocol.
988     serializeRelationship(RelationshipKind::ConformsTo, Record, Protocol);
989 
990   if (auto *ObjCInterface = dyn_cast<ObjCInterfaceRecord>(&Record)) {
991     if (!ObjCInterface->SuperClass.empty())
992       // If Record is an Objective-C interface record and it has a super class,
993       // record that Record is inherited from SuperClass.
994       serializeRelationship(RelationshipKind::InheritsFrom, Record,
995                             ObjCInterface->SuperClass);
996 
997     // Members of categories extending an interface are serialized as members of
998     // the interface.
999     for (const auto *Category : ObjCInterface->Categories) {
1000       serializeMembers(Record, Category->Ivars);
1001       serializeMembers(Record, Category->Methods);
1002       serializeMembers(Record, Category->Properties);
1003 
1004       // Surface the protocols of the category to the interface.
1005       for (const auto &Protocol : Category->Protocols)
1006         serializeRelationship(RelationshipKind::ConformsTo, Record, Protocol);
1007     }
1008   }
1009 }
1010 
1011 void SymbolGraphSerializer::visitObjCCategoryRecord(
1012     const ObjCCategoryRecord &Record) {
1013   if (!Record.IsFromExternalModule)
1014     return;
1015 
1016   // Check if the current Category' parent has been visited before, if so skip.
1017   if (!visitedCategories.contains(Record.Interface.Name)) {
1018     visitedCategories.insert(Record.Interface.Name);
1019     Object Obj;
1020     serializeObject(Obj, "identifier",
1021                     serializeIdentifier(Record, API.getLanguage()));
1022     serializeObject(Obj, "kind",
1023                     serializeSymbolKind(APIRecord::RK_ObjCCategoryModule,
1024                                         API.getLanguage()));
1025     Obj["accessLevel"] = "public";
1026     Symbols.emplace_back(std::move(Obj));
1027   }
1028 
1029   Object Relationship;
1030   Relationship["source"] = Record.USR;
1031   Relationship["target"] = Record.Interface.USR;
1032   Relationship["targetFallback"] = Record.Interface.Name;
1033   Relationship["kind"] = getRelationshipString(RelationshipKind::ExtensionTo);
1034   Relationships.emplace_back(std::move(Relationship));
1035 
1036   auto ObjCCategory = serializeAPIRecord(Record);
1037 
1038   if (!ObjCCategory)
1039     return;
1040 
1041   Symbols.emplace_back(std::move(*ObjCCategory));
1042   serializeMembers(Record, Record.Methods);
1043   serializeMembers(Record, Record.Properties);
1044 
1045   // Surface the protocols of the category to the interface.
1046   for (const auto &Protocol : Record.Protocols)
1047     serializeRelationship(RelationshipKind::ConformsTo, Record, Protocol);
1048 }
1049 
1050 void SymbolGraphSerializer::visitMacroDefinitionRecord(
1051     const MacroDefinitionRecord &Record) {
1052   auto Macro = serializeAPIRecord(Record);
1053 
1054   if (!Macro)
1055     return;
1056 
1057   Symbols.emplace_back(std::move(*Macro));
1058 }
1059 
1060 void SymbolGraphSerializer::serializeSingleRecord(const APIRecord *Record) {
1061   switch (Record->getKind()) {
1062   case APIRecord::RK_Unknown:
1063     llvm_unreachable("Records should have a known kind!");
1064   case APIRecord::RK_GlobalFunction:
1065     visitGlobalFunctionRecord(*cast<GlobalFunctionRecord>(Record));
1066     break;
1067   case APIRecord::RK_GlobalVariable:
1068     visitGlobalVariableRecord(*cast<GlobalVariableRecord>(Record));
1069     break;
1070   case APIRecord::RK_Enum:
1071     visitEnumRecord(*cast<EnumRecord>(Record));
1072     break;
1073   case APIRecord::RK_Struct:
1074     visitStructRecord(*cast<StructRecord>(Record));
1075     break;
1076   case APIRecord::RK_StaticField:
1077     visitStaticFieldRecord(*cast<StaticFieldRecord>(Record));
1078     break;
1079   case APIRecord::RK_CXXClass:
1080     visitCXXClassRecord(*cast<CXXClassRecord>(Record));
1081     break;
1082   case APIRecord::RK_ObjCInterface:
1083     visitObjCContainerRecord(*cast<ObjCInterfaceRecord>(Record));
1084     break;
1085   case APIRecord::RK_ObjCProtocol:
1086     visitObjCContainerRecord(*cast<ObjCProtocolRecord>(Record));
1087     break;
1088   case APIRecord::RK_ObjCCategory:
1089     visitObjCCategoryRecord(*cast<ObjCCategoryRecord>(Record));
1090     break;
1091   case APIRecord::RK_MacroDefinition:
1092     visitMacroDefinitionRecord(*cast<MacroDefinitionRecord>(Record));
1093     break;
1094   case APIRecord::RK_Typedef:
1095     visitTypedefRecord(*cast<TypedefRecord>(Record));
1096     break;
1097   default:
1098     if (auto Obj = serializeAPIRecord(*Record)) {
1099       Symbols.emplace_back(std::move(*Obj));
1100       auto &ParentInformation = Record->ParentInformation;
1101       if (!ParentInformation.empty())
1102         serializeRelationship(RelationshipKind::MemberOf, *Record,
1103                               *ParentInformation.ParentRecord);
1104     }
1105     break;
1106   }
1107 }
1108 
1109 void SymbolGraphSerializer::visitTypedefRecord(const TypedefRecord &Record) {
1110   // Typedefs of anonymous types have their entries unified with the underlying
1111   // type.
1112   bool ShouldDrop = Record.UnderlyingType.Name.empty();
1113   // enums declared with `NS_OPTION` have a named enum and a named typedef, with
1114   // the same name
1115   ShouldDrop |= (Record.UnderlyingType.Name == Record.Name);
1116   if (ShouldDrop)
1117     return;
1118 
1119   auto Typedef = serializeAPIRecord(Record);
1120   if (!Typedef)
1121     return;
1122 
1123   (*Typedef)["type"] = Record.UnderlyingType.USR;
1124 
1125   Symbols.emplace_back(std::move(*Typedef));
1126 }
1127 
1128 Object SymbolGraphSerializer::serialize() {
1129   traverseAPISet();
1130   return serializeCurrentGraph();
1131 }
1132 
1133 Object SymbolGraphSerializer::serializeCurrentGraph() {
1134   Object Root;
1135   serializeObject(Root, "metadata", serializeMetadata());
1136   serializeObject(Root, "module", serializeModule());
1137 
1138   Root["symbols"] = std::move(Symbols);
1139   Root["relationships"] = std::move(Relationships);
1140 
1141   return Root;
1142 }
1143 
1144 void SymbolGraphSerializer::serialize(raw_ostream &os) {
1145   Object root = serialize();
1146   if (Options.Compact)
1147     os << formatv("{0}", Value(std::move(root))) << "\n";
1148   else
1149     os << formatv("{0:2}", Value(std::move(root))) << "\n";
1150 }
1151 
1152 std::optional<Object>
1153 SymbolGraphSerializer::serializeSingleSymbolSGF(StringRef USR,
1154                                                 const APISet &API) {
1155   APIRecord *Record = API.findRecordForUSR(USR);
1156   if (!Record)
1157     return {};
1158 
1159   Object Root;
1160   APIIgnoresList EmptyIgnores;
1161   SymbolGraphSerializer Serializer(API, EmptyIgnores,
1162                                    /*Options.Compact*/ {true},
1163                                    /*ShouldRecurse*/ false);
1164   Serializer.serializeSingleRecord(Record);
1165   serializeObject(Root, "symbolGraph", Serializer.serializeCurrentGraph());
1166 
1167   Language Lang = API.getLanguage();
1168   serializeArray(Root, "parentContexts",
1169                  generateParentContexts(*Record, API, Lang));
1170 
1171   Array RelatedSymbols;
1172 
1173   for (const auto &Fragment : Record->Declaration.getFragments()) {
1174     // If we don't have a USR there isn't much we can do.
1175     if (Fragment.PreciseIdentifier.empty())
1176       continue;
1177 
1178     APIRecord *RelatedRecord = API.findRecordForUSR(Fragment.PreciseIdentifier);
1179 
1180     // If we can't find the record let's skip.
1181     if (!RelatedRecord)
1182       continue;
1183 
1184     Object RelatedSymbol;
1185     RelatedSymbol["usr"] = RelatedRecord->USR;
1186     RelatedSymbol["declarationLanguage"] = getLanguageName(Lang);
1187     // TODO: once we record this properly let's serialize it right.
1188     RelatedSymbol["accessLevel"] = "public";
1189     RelatedSymbol["filePath"] = RelatedRecord->Location.getFilename();
1190     RelatedSymbol["moduleName"] = API.ProductName;
1191     RelatedSymbol["isSystem"] = RelatedRecord->IsFromSystemHeader;
1192 
1193     serializeArray(RelatedSymbol, "parentContexts",
1194                    generateParentContexts(*RelatedRecord, API, Lang));
1195     RelatedSymbols.push_back(std::move(RelatedSymbol));
1196   }
1197 
1198   serializeArray(Root, "relatedSymbols", RelatedSymbols);
1199   return Root;
1200 }
1201