Parse now returns a vector of unique_ptrs

src/command_line.cc

@@ -67,10 +67,10 @@ struct Index_OnIndexed {
 
 // TODO: Rename TypedBidiMessageQueue to IpcTransport?
 using IpcMessageQueue = TypedBidiMessageQueue<IpcId, BaseIpcMessage>;
-using Index_DoIndexQueue = ThreadedQueue<std::unique_ptr<Index_DoIndex>>;
-using Index_DoIdMapQueue = ThreadedQueue<std::unique_ptr<Index_DoIdMap>>;
-using Index_OnIdMappedQueue = ThreadedQueue<std::unique_ptr<Index_OnIdMapped>>;
-using Index_OnIndexedQueue = ThreadedQueue<std::unique_ptr<Index_OnIndexed>>;
+using Index_DoIndexQueue = ThreadedQueue<Index_DoIndex>;
+using Index_DoIdMapQueue = ThreadedQueue<Index_DoIdMap>;
+using Index_OnIdMappedQueue = ThreadedQueue<Index_OnIdMapped>;
+using Index_OnIndexedQueue = ThreadedQueue<Index_OnIndexed>;
 
 template<typename TMessage>
 void SendMessage(IpcMessageQueue& t, MessageQueue* destination, TMessage& message) {
@@ -291,44 +291,57 @@ void WriteToCache(std::string filename, IndexedFile& file) {
 
 bool IndexMain_DoIndex(Index_DoIndexQueue* queue_do_index,
   Index_DoIdMapQueue* queue_do_id_map) {
-  optional<std::unique_ptr<Index_DoIndex>> opt_index_request = queue_do_index->TryDequeue();
-  if (!opt_index_request)
+  optional<Index_DoIndex> index_request = queue_do_index->TryDequeue();
+  if (!index_request)
     return false;
-  std::unique_ptr<Index_DoIndex> index_request = std::move(opt_index_request.value());
 
   Timer time;
 
-  std::unique_ptr<IndexedFile> old_index = LoadCachedFile(index_request->path);
-  time.ResetAndPrint("Loading cached index");
+
 
 
   // If the index update is an import, then we will load the previous index
   // into memory if we have a previous index. After that, we dispatch an
   // update request to get the latest version.
-  if (old_index && index_request->type == Index_DoIndex::Type::Import) {
-    auto response = MakeUnique<Index_DoIdMap>(nullptr /*previous*/, std::move(old_index) /*current*/);
+  if (index_request->type == Index_DoIndex::Type::Import) {
+    index_request->type = Index_DoIndex::Type::Update;
+    std::unique_ptr<IndexedFile> old_index = LoadCachedFile(index_request->path);
+    time.ResetAndPrint("Loading cached index");
+
+    // If import fails just do a standard update.
+    if (old_index) {
+      Index_DoIdMap response(nullptr, std::move(old_index));
       queue_do_id_map->Enqueue(std::move(response));
 
-    index_request->type = Index_DoIndex::Type::Update;
-    queue_do_index->Enqueue(std::move(index_request));
+      queue_do_index->Enqueue(std::move(*index_request));
+      return true;
     }
-  else {
+  }
+
   // Parse request and send a response.
   std::cerr << "Parsing file " << index_request->path << " with args "
     << Join(index_request->args, ", ") << std::endl;
 
   // TODO: parse should return unique_ptr. Then we can eliminate copy below. Make sure to not
   // reuse moved pointer in WriteToCache if we do so.
-    IndexedFile current_index = Parse(index_request->path, index_request->args);
-
+  std::vector<std::unique_ptr<IndexedFile>> indexes = Parse(index_request->path, index_request->args);
   time.ResetAndPrint("Parsing/indexing");
 
-    auto response = MakeUnique<Index_DoIdMap>(std::move(old_index) /*previous*/, MakeUnique<IndexedFile>(current_index) /*current*/);
-    queue_do_id_map->Enqueue(std::move(response));
+  for (auto& current_index : indexes) {
+    std::unique_ptr<IndexedFile> old_index = LoadCachedFile(current_index->path);
+    time.ResetAndPrint("Loading cached index");
+
+    // TODO: Cache to disk on a separate thread. Maybe we do the cache after we
+    // have imported the index (so the import pipeline has five stages instead
+    // of the current 4).
 
     // Cache file so we can diff it later.
-    WriteToCache(index_request->path, current_index);
+    WriteToCache(index_request->path, *current_index);
     time.ResetAndPrint("Cache index update to disk");
+
+    // Send response to create id map.
+    Index_DoIdMap response(std::move(old_index), std::move(current_index));
+    queue_do_id_map->Enqueue(std::move(response));
   }
 
   return true;
@@ -336,16 +349,15 @@ bool IndexMain_DoIndex(Index_DoIndexQueue* queue_do_index,
 
 bool IndexMain_DoCreateIndexUpdate(Index_OnIdMappedQueue* queue_on_id_mapped,
   Index_OnIndexedQueue* queue_on_indexed) {
-  optional<std::unique_ptr<Index_OnIdMapped>> opt_response = queue_on_id_mapped->TryDequeue();
-  if (!opt_response)
+  optional<Index_OnIdMapped> response = queue_on_id_mapped->TryDequeue();
+  if (!response)
     return false;
-  std::unique_ptr<Index_OnIdMapped> response = std::move(opt_response.value());
 
   Timer time;
   IndexUpdate update = IndexUpdate::CreateDelta(response->previous_id_map.get(), response->current_id_map.get(),
     response->previous_index.get(), response->current_index.get());
   time.ResetAndPrint("Creating delta IndexUpdate");
-  auto reply = MakeUnique<Index_OnIndexed>(update);
+  Index_OnIndexed reply(update);
   queue_on_indexed->Enqueue(std::move(reply));
   time.ResetAndPrint("Sending update to server");
 
@@ -370,15 +382,9 @@ void IndexMain(Index_DoIndexQueue* queue_do_index,
 }
 
 QueryableFile* FindFile(QueryableDatabase* db, const std::string& filename) {
-  // std::cerr << "Wanted file " << msg->document << std::endl;
-  // TODO: hashmap lookup.
-  for (auto& file : db->files) {
-    // std::cerr << " - Have file " << file.file_id << std::endl;
-    if (file.def.usr == filename) {
-      //std::cerr << "Found file " << filename << std::endl;
-      return &file;
-    }
-  }
+  auto it = db->usr_to_symbol.find(filename);
+  if (it != db->usr_to_symbol.end())
+    return &db->files[it->second.idx];
 
   std::cerr << "Unable to find file " << filename << std::endl;
   return nullptr;
@@ -635,9 +641,9 @@ void QueryDbMainLoop(
           << "] Dispatching index request for file " << filepath
           << std::endl;
 
-        auto request = MakeUnique<Index_DoIndex>(Index_DoIndex::Type::Import);
-        request->path = filepath;
-        request->args = entry.args;
+        Index_DoIndex request(Index_DoIndex::Type::Import);
+        request.path = filepath;
+        request.args = entry.args;
         queue_do_index->Enqueue(std::move(request));
       }
       std::cerr << "Done" << std::endl;
@@ -932,29 +938,30 @@ void QueryDbMainLoop(
 
 
   while (true) {
-    optional<std::unique_ptr<Index_DoIdMap>> opt_request = queue_do_id_map->TryDequeue();
-    if (!opt_request)
+    optional<Index_DoIdMap> request = queue_do_id_map->TryDequeue();
+    if (!request)
       break;
-    std::unique_ptr<Index_DoIdMap> request = std::move(opt_request.value());
 
-    auto response = MakeUnique<Index_OnIdMapped>();
+
+    Index_OnIdMapped response;
     Timer time;
     if (request->previous) {
-      response->previous_id_map = MakeUnique<IdMap>(db, request->previous->id_cache);
-      response->previous_index = std::move(request->previous);
+      response.previous_id_map = MakeUnique<IdMap>(db, request->previous->id_cache);
+      response.previous_index = std::move(request->previous);
     }
-    response->current_id_map = MakeUnique<IdMap>(db, request->current->id_cache);
-    response->current_index = std::move(request->current);
+
+    assert(request->current);
+    response.current_id_map = MakeUnique<IdMap>(db, request->current->id_cache);
+    response.current_index = std::move(request->current);
     time.ResetAndPrint("Create IdMap");
 
     queue_on_id_mapped->Enqueue(std::move(response));
   }
 
   while (true) {
-    optional<std::unique_ptr<Index_OnIndexed>> opt_response = queue_on_indexed->TryDequeue();
-    if (!opt_response)
+    optional<Index_OnIndexed> response = queue_on_indexed->TryDequeue();
+    if (!response)
       break;
-    std::unique_ptr<Index_OnIndexed> response = std::move(opt_response.value());
 
     Timer time;
     db->ApplyIndexUpdate(&response->update);

src/file_consumer.cc

@@ -0,0 +1,24 @@
+#include "file_consumer.h"
+
+FileConsumer::FileConsumer(SharedState* shared_state) : shared_(shared_state) {}
+
+void FileConsumer::ClearOwnership() {
+  for (auto& entry : local_)
+    entry.second = Ownership::DoesNotOwn;
+}
+
+bool FileConsumer::DoesOwnFile(const std::string& file) {
+  // Try to find cached local result.
+  auto it = local_.find(file);
+  if (it != local_.end())
+    return it->second == Ownership::Owns;
+
+  // No result in local; we need to query global.
+  bool did_insert = false;
+  {
+    std::lock_guard<std::mutex> lock(shared_->muetx);
+    did_insert = shared_->files.insert(file).second;
+  }
+  local_[file] = did_insert ? Ownership::Owns : Ownership::DoesNotOwn;
+  return did_insert;
+}

src/file_consumer.h

@@ -0,0 +1,37 @@
+#pragma once
+
+#include <mutex>
+#include <unordered_set>
+#include <unordered_map>
+
+// FileConsumer is used by the indexer. When it encouters a file, it tries to
+// take ownership over it. If the indexer has ownership over a file, it will
+// produce an index, otherwise, it will emit nothing for that declarations
+// and references coming from that file.
+//
+// The indexer does this because header files do not have their own translation
+// units but we still want to index them.
+struct FileConsumer {
+  struct SharedState {
+    mutable std::unordered_set<std::string> files;
+    mutable std::mutex muetx;
+  };
+
+  FileConsumer(SharedState* shared_state);
+
+  // Returns true if this instance owns given |file|. This will also attempt to
+  // take ownership over |file|.
+  bool DoesOwnFile(const std::string& file);
+
+  // Clear all ownership state.
+  void ClearOwnership();
+
+ private:
+  enum class Ownership {
+    Owns,
+    DoesNotOwn
+  };
+
+  std::unordered_map<std::string, Ownership> local_;
+  SharedState* shared_;
+};

src/indexer.cc

@@ -1263,12 +1263,7 @@ void indexEntityReference(CXClientData client_data,
   }
 }
 
-IndexedFile Parse(std::string filename,
-                  std::vector<std::string> args,
-                  bool dump_ast) {
-  // TODO: We are currently emitting too much information for things not in the main file. If we're
-  // not in the main file, we should only emit references.
-
+std::vector<std::unique_ptr<IndexedFile>> Parse(std::string filename, std::vector<std::string> args, bool dump_ast) {
   clang_enableStackTraces();
   clang_toggleCrashRecovery(1);
 
@@ -1300,9 +1295,9 @@ IndexedFile Parse(std::string filename,
       */
   };
 
-  IndexedFile db(filename);
+  auto db = MakeUnique<IndexedFile>(filename);
   NamespaceHelper ns;
-  IndexParam param(&db, &ns);
+  IndexParam param(db.get(), &ns);
 
   std::cerr << "!! [START] Indexing " << filename << std::endl;
   clang_indexTranslationUnit(index_action, &param, callbacks, sizeof(callbacks),
@@ -1311,5 +1306,7 @@ IndexedFile Parse(std::string filename,
   std::cerr << "!! [END] Indexing " << filename << std::endl;
   clang_IndexAction_dispose(index_action);
 
-  return db;
+  std::vector<std::unique_ptr<IndexedFile>> result;
+  result.emplace_back(std::move(db));
+  return std::move(result);
 }

src/indexer.h

@@ -448,6 +448,4 @@ struct IndexedFile {
   std::string ToString();
 };
 
-IndexedFile Parse(std::string filename,
-                  std::vector<std::string> args,
-                  bool dump_ast = false);
+std::vector<std::unique_ptr<IndexedFile>> Parse(std::string filename, std::vector<std::string> args, bool dump_ast = false);

src/project.cc

@@ -23,10 +23,7 @@ std::vector<CompilationEntry> LoadFromDirectoryListing(const std::string& projec
 
   std::vector<std::string> files = GetFilesInFolder(project_directory, true /*recursive*/, true /*add_folder_to_path*/);
   for (const std::string& file : files) {
-    if (EndsWith(file, ".cc") || EndsWith(file, ".cpp") ||
-      EndsWith(file, ".c") || EndsWith(file, ".h") ||
-      EndsWith(file, ".hpp")) {
-
+    if (EndsWith(file, ".cc") || EndsWith(file, ".cpp") || EndsWith(file, ".c") || EndsWith(file, ".h")) {
       CompilationEntry entry;
       entry.filename = NormalizePath(file);
       entry.args = args;

src/test.cc

@@ -72,20 +72,20 @@ void DiffDocuments(std::string path, rapidjson::Document& expected, rapidjson::D
 
   if (actual_output.size() > len) {
     std::cout << "Additional output in actual:" << std::endl;
-    for (int i = len; i < actual_output.size(); ++i)
+    for (size_t i = len; i < actual_output.size(); ++i)
       std::cout << "  " << actual_output[i] << std::endl;
   }
 
   if (expected_output.size() > len) {
     std::cout << "Additional output in expected:" << std::endl;
-    for (int i = len; i < expected_output.size(); ++i)
+    for (size_t i = len; i < expected_output.size(); ++i)
       std::cout << "  " << expected_output[i] << std::endl;
   }
 }
 
-void VerifySerializeToFrom(IndexedFile& file) {
-  std::string expected = file.ToString();
-  std::string actual = Deserialize("--.cc", Serialize(file)).value().ToString();
+void VerifySerializeToFrom(IndexedFile* file) {
+  std::string expected = file->ToString();
+  std::string actual = Deserialize("--.cc", Serialize(*file)).value().ToString();
   if (expected != actual) {
     std::cerr << "Serialization failure" << std::endl;;
     assert(false);
@@ -123,7 +123,7 @@ void RunTests() {
 
     // Run test.
     std::cout << "[START] " << path << std::endl;
-    IndexedFile db = Parse(path, {
+    std::vector<std::unique_ptr<IndexedFile>> dbs = Parse(path, {
         "-xc++",
         "-std=c++11",
         "-IC:/Users/jacob/Desktop/superindex/indexer/third_party/",
@@ -131,8 +131,23 @@ void RunTests() {
         "-IC:/Users/jacob/Desktop/superindex/indexer/third_party/rapidjson/include",
         "-IC:/Users/jacob/Desktop/superindex/indexer/src"
       }, false /*dump_ast*/);
-    VerifySerializeToFrom(db);
-    std::string actual_output = db.ToString();
+
+    // TODO: Supporting tests for more than just primary indexed file.
+
+    // Find primary file.
+    std::unique_ptr<IndexedFile> db;
+    for (auto& i : dbs) {
+      if (i->path == path) {
+        db = std::move(i);
+        break;
+      }
+    }
+
+    // TODO: Always pass IndexedFile by pointer, ie, search and remove all IndexedFile& refs.
+    // TODO: Rename IndexedFile to IndexFile
+
+    VerifySerializeToFrom(db.get());
+    std::string actual_output = db->ToString();
 
     rapidjson::Document actual;
     actual.Parse(actual_output.c_str());