[lldb] Teach LLDB to load gmodules from CAS

Author: cachemeifyoucan

clang/test/ClangScanDeps/gmodules.c

@@ -218,6 +218,7 @@ int main(void) {
   struct Right _right;
   struct Top _top;
   struct Prefix _prefix;
+  return 0;
 }
 
 //--- cas-config.template

lldb/include/lldb/Core/ModuleList.h

@@ -528,6 +528,11 @@ class ModuleList {
                   llvm::SmallVectorImpl<lldb::ModuleSP> *old_modules,
                   bool *did_create_ptr, bool always_create = false);
 
+  static Status GetSharedModuleFromCAS(ConstString module_name,
+                                       const char *cas_id, FileSpec cu_path,
+                                       ModuleSpec &module_spec,
+                                       lldb::ModuleSP &module_sp);
+
   static bool RemoveSharedModule(lldb::ModuleSP &module_sp);
 
   static void FindSharedModules(const ModuleSpec &module_spec,

lldb/source/Core/ModuleList.cpp

@@ -26,13 +26,16 @@
 #include "lldb/Utility/Log.h"
 #include "lldb/Utility/UUID.h"
 #include "lldb/lldb-defines.h"
+#include "llvm/Support/FileUtilities.h"
 
 #if defined(_WIN32)
 #include "lldb/Host/windows/PosixApi.h"
 #endif
 
 #include "clang/Driver/Driver.h"
 #include "llvm/ADT/StringRef.h"
+#include "llvm/CAS/CASConfiguration.h"
+#include "llvm/CAS/ObjectStore.h"
 #include "llvm/Support/FileSystem.h"
 #include "llvm/Support/Threading.h"
 #include "llvm/Support/raw_ostream.h"
@@ -1050,8 +1053,13 @@ namespace {
 struct SharedModuleListInfo {
   ModuleList module_list;
   ModuleListProperties module_list_properties;
+  std::shared_ptr<llvm::cas::ObjectStore> cas_object_store;
+  llvm::StringMap<FileSpec> loaded_cas_files;
+  std::vector<std::unique_ptr<llvm::FileRemover>> module_file_removers;
+  std::mutex shared_lock;
 };
 }
+
 static SharedModuleListInfo &GetSharedModuleListInfo()
 {
   static SharedModuleListInfo *g_shared_module_list_info = nullptr;
@@ -1070,6 +1078,45 @@ static ModuleList &GetSharedModuleList() {
   return GetSharedModuleListInfo().module_list;
 }
 
+static std::shared_ptr<llvm::cas::ObjectStore>
+GetOrCreateCASStorage(FileSpec CandidateConfigSearchPath) {
+  auto &shared_module_list = GetSharedModuleListInfo();
+  if (shared_module_list.cas_object_store)
+    return shared_module_list.cas_object_store;
+
+  std::scoped_lock<std::mutex> lock(shared_module_list.shared_lock);
+  // Config CAS from properties.
+  llvm::cas::CASConfiguration cas_config;
+  cas_config.CASPath =
+      ModuleList::GetGlobalModuleListProperties().GetCASOnDiskPath().GetPath();
+  cas_config.PluginPath =
+      ModuleList::GetGlobalModuleListProperties().GetCASPluginPath().GetPath();
+  cas_config.PluginOptions =
+      ModuleList::GetGlobalModuleListProperties().GetCASPluginOptions();
+
+  if (!cas_config.CASPath.empty()) {
+    if (auto maybe_cas = cas_config.createDatabases()) {
+      shared_module_list.cas_object_store = std::move(maybe_cas->first);
+      return shared_module_list.cas_object_store;
+    } else
+      llvm::consumeError(maybe_cas.takeError());
+  }
+
+  // Try search from candidiate path.
+  auto search_config = llvm::cas::CASConfiguration::createFromSearchConfigFile(
+      CandidateConfigSearchPath.GetPath());
+  if (!search_config)
+    return nullptr;
+
+  if (auto maybe_cas = search_config->second.createDatabases()) {
+    shared_module_list.cas_object_store = std::move(maybe_cas->first);
+    return shared_module_list.cas_object_store;
+  } else
+    llvm::consumeError(maybe_cas.takeError());
+
+  return nullptr;
+}
+
 ModuleListProperties &ModuleList::GetGlobalModuleListProperties() {
   return GetSharedModuleListInfo().module_list_properties;
 }
@@ -1342,6 +1389,66 @@ ModuleList::GetSharedModule(const ModuleSpec &module_spec, ModuleSP &module_sp,
   return error;
 }
 
+static Status loadModuleFromCAS(ConstString module_name, const char *cas_id,
+                                FileSpec cu_path, ModuleSpec &module_spec) {
+  auto cas = GetOrCreateCASStorage(cu_path);
+  if (!cas)
+    return Status::FromErrorString("CAS is not available");
+
+  auto &shared_module_list = GetSharedModuleListInfo();
+  std::scoped_lock<std::mutex> lock(shared_module_list.shared_lock);
+  auto result = shared_module_list.loaded_cas_files.find(cas_id);
+  if (result != shared_module_list.loaded_cas_files.end()) {
+    module_spec.GetFileSpec() = result->second;
+    return Status();
+  }
+
+  auto id = cas->parseID(cas_id) ;
+  if (!id)
+    return Status::FromError(id.takeError());
+
+  auto module_proxy = cas->getProxy(*id);
+  if (!module_proxy)
+    return Status::FromError(module_proxy.takeError());
+
+  // Create a temporary file that holds the module info.
+  // TODO: remove temporary file when finished.
+  llvm::SmallString<0> name;
+  int fd;
+  auto ec =
+      llvm::sys::fs::createTemporaryFile("lldb-module-cas", "pcm", fd, name);
+  if (ec || fd <= 0)
+    return Status::FromErrorString(
+        "Could not create temporary file for module");
+
+  NativeFile file(fd, File::eOpenOptionWriteOnly, true);
+  size_t num_bytes = module_proxy->getData().size();
+  file.Write(module_proxy->getData().data(), num_bytes);
+
+  Log *log = GetLog(LLDBLog::Modules);
+  if (log != nullptr)
+    LLDB_LOGF(log, "loading module '%s' using CASID '%s' from temp file: %s",
+              module_name.AsCString(), cas_id, name.c_str());
+
+  module_spec.GetFileSpec().SetFile(name, FileSpec::Style::native);
+  shared_module_list.loaded_cas_files.try_emplace(cas_id,
+                                                  module_spec.GetFileSpec());
+
+  return Status();
+}
+
+Status ModuleList::GetSharedModuleFromCAS(ConstString module_name,
+                                          const char *cas_id, FileSpec cu_path,
+                                          ModuleSpec &module_spec,
+                                          lldb::ModuleSP &module_sp) {
+  Status load_result =
+      loadModuleFromCAS(module_name, cas_id, cu_path, module_spec);
+  if (load_result.Fail())
+    return load_result;
+
+  return GetSharedModule(module_spec, module_sp, nullptr, nullptr, nullptr);
+}
+
 bool ModuleList::RemoveSharedModule(lldb::ModuleSP &module_sp) {
   return GetSharedModuleList().Remove(module_sp);
 }

lldb/source/Plugins/SymbolFile/DWARF/SymbolFileDWARF.cpp

@@ -1998,6 +1998,17 @@ void SymbolFileDWARF::UpdateExternalModuleListIfNeeded() {
       continue;
 
     ModuleSpec dwo_module_spec;
+
+    dwo_module_spec.GetArchitecture() =
+        m_objfile_sp->GetModule()->GetArchitecture();
+
+    // Try load from CAS, if loaded, continue to next one.
+    if (ModuleList::GetSharedModuleFromCAS(const_name, dwo_path,
+                                           GetObjectFile()->GetFileSpec(),
+                                           dwo_module_spec, module_sp)
+            .Success())
+      continue;
+
     dwo_module_spec.GetFileSpec().SetFile(dwo_path, FileSpec::Style::native);
     if (dwo_module_spec.GetFileSpec().IsRelative()) {
       const char *comp_dir =
@@ -2009,8 +2020,6 @@ void SymbolFileDWARF::UpdateExternalModuleListIfNeeded() {
         dwo_module_spec.GetFileSpec().AppendPathComponent(dwo_path);
       }
     }
-    dwo_module_spec.GetArchitecture() =
-        m_objfile_sp->GetModule()->GetArchitecture();
 
     // When LLDB loads "external" modules it looks at the presence of
     // DW_AT_dwo_name. However, when the already created module