Introduce ExternalTypeRefInfoProvider

LLDB would like to cache typeref information to accelerate finding type
information. This patch adds an optional interface that allows of
typeref to register and provider field descriptor information for faster
lookups.

Author: augusto2112

include/swift/Reflection/ReflectionContext.h

@@ -211,9 +211,10 @@ class ReflectionContext
     uint32_t ThreadPort;
   };
 
-  explicit ReflectionContext(std::shared_ptr<MemoryReader> reader)
-    : super(std::move(reader), *this)
-  {}
+  explicit ReflectionContext(
+      std::shared_ptr<MemoryReader> reader,
+      remote::ExternalTypeRefCacheProvider *provider = nullptr)
+      : super(std::move(reader), *this, provider) {}
 
   ReflectionContext(const ReflectionContext &other) = delete;
   ReflectionContext &operator=(const ReflectionContext &other) = delete;
@@ -228,7 +229,8 @@ class ReflectionContext
   }
 
   template <typename T>
-  bool readMachOSections(
+  llvm::Optional<uint64_t>
+  readMachOSections(
       RemoteAddress ImageStart,
       llvm::SmallVector<llvm::StringRef, 1> PotentialModuleNames = {}) {
     auto Buf =
@@ -350,7 +352,7 @@ class ReflectionContext
                            {MPEnumMdSec.first, MPEnumMdSec.second},
                            PotentialModuleNames};
 
-    this->addReflectionInfo(info);
+    auto InfoID = this->addReflectionInfo(info);
 
     // Find the __DATA segment.
     for (unsigned I = 0; I < NumCommands; ++I) {
@@ -374,10 +376,10 @@ class ReflectionContext
 
     savedBuffers.push_back(std::move(Buf));
     savedBuffers.push_back(std::move(Sections));
-    return true;
+    return InfoID;
   }
 
-  bool readPECOFFSections(
+  llvm::Optional<uint64_t> readPECOFFSections(
       RemoteAddress ImageStart,
       llvm::SmallVector<llvm::StringRef, 1> PotentialModuleNames = {}) {
     auto DOSHdrBuf = this->getReader().readBytes(
@@ -477,11 +479,10 @@ class ReflectionContext
                            {ConformMdSec.first, ConformMdSec.second},
                            {MPEnumMdSec.first, MPEnumMdSec.second},
                            PotentialModuleNames};
-    this->addReflectionInfo(Info);
-    return true;
+    return this->addReflectionInfo(Info);
   }
 
-  bool readPECOFF(RemoteAddress ImageStart,
+  llvm::Optional<uint64_t> readPECOFF(RemoteAddress ImageStart,
                   llvm::SmallVector<llvm::StringRef, 1> PotentialModuleNames = {}) {
     auto Buf = this->getReader().readBytes(ImageStart,
                                            sizeof(llvm::object::dos_header));
@@ -505,7 +506,7 @@ class ReflectionContext
   }
 
   template <typename T>
-  bool readELFSections(
+  llvm::Optional<uint64_t> readELFSections(
       RemoteAddress ImageStart,
       llvm::Optional<llvm::sys::MemoryBlock> FileBuffer,
       llvm::SmallVector<llvm::StringRef, 1> PotentialModuleNames = {}) {
@@ -673,8 +674,7 @@ class ReflectionContext
                            {MPEnumMdSec.first, MPEnumMdSec.second},
                            PotentialModuleNames};
 
-    this->addReflectionInfo(info);
-    return true;
+    return this->addReflectionInfo(info);
   }
 
   /// Parses metadata information from an ELF image. Because the Section
@@ -693,22 +693,22 @@ class ReflectionContext
   ///     instance's memory reader.
   ///
   /// \return
-  ///     /b True if the metadata information was parsed successfully,
-  ///     /b false otherwise.
-  bool
+  ///     \b  The newly added reflection info ID if successful,
+  ///     \b llvm::None otherwise.
+  llvm::Optional<uint64_t>
   readELF(RemoteAddress ImageStart,
           llvm::Optional<llvm::sys::MemoryBlock> FileBuffer,
           llvm::SmallVector<llvm::StringRef, 1> PotentialModuleNames = {}) {
     auto Buf =
         this->getReader().readBytes(ImageStart, sizeof(llvm::ELF::Elf64_Ehdr));
     if (!Buf)
-      return false;
+      return llvm::None;
 
     // Read the header.
     auto Hdr = reinterpret_cast<const llvm::ELF::Elf64_Ehdr *>(Buf.get());
 
     if (!Hdr->checkMagic())
-      return false;
+      return llvm::None;
 
     // Check if we have a ELFCLASS32 or ELFCLASS64
     unsigned char FileClass = Hdr->getFileClass();
@@ -719,11 +719,11 @@ class ReflectionContext
       return readELFSections<ELFTraits<llvm::ELF::ELFCLASS32>>(
           ImageStart, FileBuffer, PotentialModuleNames);
     } else {
-      return false;
+      return llvm::None;
     }
   }
 
-  bool
+  llvm::Optional<uint64_t>
   addImage(RemoteAddress ImageStart,
            llvm::SmallVector<llvm::StringRef, 1> PotentialModuleNames = {}) {
     // Read the first few bytes to look for a magic header.
@@ -761,7 +761,7 @@ class ReflectionContext
     }
 
     // We don't recognize the format.
-    return false;
+    return llvm::None;
   }
 
   /// Adds an image using the FindSection closure to find the swift metadata
@@ -770,9 +770,9 @@ class ReflectionContext
   ///     of freeing the memory buffer in the RemoteRef return value.
   ///     process.
   /// \return
-  ///     \b True if any of the reflection sections were registered,
-  ///     \b false otherwise.
-  bool
+  ///     \b  The newly added reflection info ID if successful,
+  ///     \b llvm::None otherwise.
+  llvm::Optional<uint64_t>
   addImage(llvm::function_ref<
                std::pair<RemoteRef<void>, uint64_t>(ReflectionSectionKind)>
                FindSection,
@@ -798,7 +798,7 @@ class ReflectionContext
 
     // If we didn't find any sections, return.
     if (llvm::all_of(Pairs, [](const auto &Pair) { return !Pair.first; }))
-      return false;
+      return {};
 
     ReflectionInfo Info = {{Pairs[0].first, Pairs[0].second},
                            {Pairs[1].first, Pairs[1].second},
@@ -809,12 +809,11 @@ class ReflectionContext
                            {Pairs[6].first, Pairs[6].second},
                            {Pairs[7].first, Pairs[7].second},
                            PotentialModuleNames};
-    this->addReflectionInfo(Info);
-    return true;
+    return addReflectionInfo(Info);
   }
 
-  void addReflectionInfo(ReflectionInfo I) {
-    getBuilder().addReflectionInfo(I);
+  llvm::Optional<uint64_t> addReflectionInfo(ReflectionInfo I) {
+    return getBuilder().addReflectionInfo(I);
   }
 
   bool ownsObject(RemoteAddress ObjectAddress) {

include/swift/Reflection/TypeRefBuilder.h

@@ -18,6 +18,7 @@
 #ifndef SWIFT_REFLECTION_TYPEREFBUILDER_H
 #define SWIFT_REFLECTION_TYPEREFBUILDER_H
 
+#include "swift/Remote/ExternalTypeRefCacheProvider.h"
 #include "swift/Remote/MetadataReader.h"
 #include "swift/Reflection/MetadataSourceBuilder.h"
 #include "swift/Reflection/Records.h"
@@ -426,6 +427,8 @@ class TypeRefBuilder {
   TypeConverter TC;
   MetadataSourceBuilder MSB;
 
+  remote::ExternalTypeRefCacheProvider *ExternalTypeRefCache = nullptr;
+
 #define TYPEREF(Id, Parent) \
   std::unordered_map<TypeRefID, const Id##TypeRef *, \
                      TypeRefID::Hash, TypeRefID::Equal> Id##TypeRefs;
@@ -915,10 +918,13 @@ class TypeRefBuilder {
   /// Parsing reflection metadata
   ///
 
-  void addReflectionInfo(ReflectionInfo I) {
+  /// Add the ReflectionInfo and return a unique ID for the reflection image
+  /// added. Since we only add reflection infos, the ID can be its index.
+  llvm::Optional<uint64_t> addReflectionInfo(ReflectionInfo I) {
     ReflectionInfos.push_back(I);
+    return ReflectionInfos.size() - 1;
   }
-  
+
   const std::vector<ReflectionInfo> &getReflectionInfos() {
     return ReflectionInfos;
   }
@@ -943,6 +949,9 @@ class TypeRefBuilder {
   llvm::Optional<RemoteRef<FieldDescriptor>>
   findFieldDescriptorAtIndex(size_t Index, const std::string &MangledName);
 
+  llvm::Optional<RemoteRef<FieldDescriptor>>
+  getFieldDescriptorFromProvider(const std::string &MangledName);
+
 public:
   RemoteRef<char> readTypeRef(uint64_t remoteAddr);
 
@@ -983,8 +992,9 @@ class TypeRefBuilder {
 
 public:
   template<typename Runtime>
-  TypeRefBuilder(remote::MetadataReader<Runtime, TypeRefBuilder> &reader)
-    : TC(*this),
+  TypeRefBuilder(remote::MetadataReader<Runtime, TypeRefBuilder> &reader,
+                 remote::ExternalTypeRefCacheProvider *provider = nullptr)
+      : TC(*this), ExternalTypeRefCache(provider),
       PointerSize(sizeof(typename Runtime::StoredPointer)),
       TypeRefDemangler(
       [this, &reader](RemoteRef<char> string, bool useOpaqueTypeSymbolicReferences) -> Demangle::Node * {

include/swift/Remote/ExternalTypeRefCacheProvider.h

@@ -0,0 +1,74 @@
+//===--- ExternalTypeRefCacheProvider.h - Abstract access to external caches of
+//typeref ------*- C++ -*-===//
+//
+// This source file is part of the Swift.org open source project
+//
+// Copyright (c) 2022 Apple Inc. and the Swift project authors
+// Licensed under Apache License v2.0 with Runtime Library Exception
+//
+// See https://swift.org/LICENSE.txt for license information
+// See https://swift.org/CONTRIBUTORS.txt for the list of Swift project authors
+//
+//===----------------------------------------------------------------------===//
+//
+//  This file declares an abstract interface for external cache providers of
+//  typeref information
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef SWIFT_REMOTE_EXTERNALTYPEREFCACHEPROVIDER_H
+#define SWIFT_REMOTE_EXTERNALTYPEREFCACHEPROVIDER_H
+
+#include "llvm/ADT/ArrayRef.h"
+#include "llvm/ADT/StringRef.h"
+#include "llvm/ADT/Optional.h"
+
+#include <string>
+
+namespace swift {
+namespace reflection {
+
+template <typename T>
+class ReflectionSection;
+class FieldDescriptorIterator;
+using FieldSection = ReflectionSection<FieldDescriptorIterator>;
+}
+
+namespace remote {
+/// A struct with the information requited to locate a specific field
+/// descriptor.
+struct FieldDescriptorLocator {
+  /// The reflection info ID the field descriptor belongs to.
+  uint64_t InfoID;
+
+  /// The offset of the field descriptor in the FieldSection buffer.
+  uint64_t Offset;
+};
+
+/// An abstract interface for providing external type layout information.
+struct ExternalTypeRefCacheProvider {
+  virtual ~ExternalTypeRefCacheProvider() = default;
+
+  /// Cache the field descriptors of a reflection info with a given id with
+  /// their corresponding mangled names. The amount of field descriptors and
+  /// mangled names must be the same. If a field descriptor does not have a
+  /// mangled name a corresponding empty string must be in the mangled_names
+  /// array.
+  virtual void cacheFieldDescriptors(
+      uint64_t InfoID,
+      const swift::reflection::FieldSection &FieldDescriptors,
+      const llvm::ArrayRef<std::string> &MangledNames) = 0;
+
+  /// Retrieve a pair representing the reflection info id and the offset of a
+  /// field descriptor in the field section buffer, if available.
+  virtual llvm::Optional<FieldDescriptorLocator>
+  getFieldDescriptorLocator(const std::string &Name) = 0;
+
+  /// Returns whether the reflection info with the corresponding ID has been
+  /// cached already.
+  virtual bool isReflectionInfoCached(uint64_t InfoID) = 0;
+};
+
+} // namespace remote
+} // namespace swift
+#endif