Initial commit of sync engine code to browser/sync....

chrome/browser/sync/engine/syncapi.cc

Index: chrome/browser/sync/engine/syncapi.cc
===================================================================
--- chrome/browser/sync/engine/syncapi.cc	(revision 0)
+++ chrome/browser/sync/engine/syncapi.cc	(revision 0)
@@ -0,0 +1,1565 @@
+// Copyright (c) 2006-2009 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "chrome/browser/sync/engine/syncapi.h"
+
+#if defined(OS_WINDOWS)
+#include <windows.h>
+#include <iphlpapi.h>
+#endif
+
+#include <iomanip>
+#include <list>
+#include <string>
+#include <vector>
+
+#include "base/at_exit.h"
+#include "base/basictypes.h"
+#include "base/scoped_ptr.h"
+#include "base/string_util.h"
+#include "chrome/browser/sync/engine/all_status.h"
+#include "chrome/browser/sync/engine/auth_watcher.h"
+#include "chrome/browser/sync/engine/change_reorder_buffer.h"
+#include "chrome/browser/sync/engine/client_command_channel.h"
+#include "chrome/browser/sync/engine/model_safe_worker.h"
+#include "chrome/browser/sync/engine/net/gaia_authenticator.h"
+#include "chrome/browser/sync/engine/net/server_connection_manager.h"
+#include "chrome/browser/sync/engine/net/syncapi_server_connection_manager.h"
+#include "chrome/browser/sync/engine/syncer.h"
+#include "chrome/browser/sync/engine/syncer_thread.h"
+#include "chrome/browser/sync/notifier/listener/talk_mediator.h"
+#include "chrome/browser/sync/notifier/listener/talk_mediator_impl.h"
+#include "chrome/browser/sync/protocol/service_constants.h"
+#include "chrome/browser/sync/syncable/directory_manager.h"
+#include "chrome/browser/sync/syncable/syncable.h"
+#include "chrome/browser/sync/util/character_set_converters.h"
+#include "chrome/browser/sync/util/closure.h"
+#include "chrome/browser/sync/util/crypto_helpers.h"
+#include "chrome/browser/sync/util/event_sys.h"
+#include "chrome/browser/sync/util/path_helpers.h"
+#include "chrome/browser/sync/util/pthread_helpers.h"
+#include "chrome/browser/sync/util/user_settings.h"
+#include "googleurl/src/gurl.h"
+
+using browser_sync::AllStatus;
+using browser_sync::AllStatusEvent;
+using browser_sync::AuthWatcher;
+using browser_sync::AuthWatcherEvent;
+using browser_sync::ClientCommandChannel;
+using browser_sync::Syncer;
+using browser_sync::SyncerEvent;
+using browser_sync::SyncerStatus;
+using browser_sync::SyncerThread;
+using browser_sync::UserSettings;
+using browser_sync::TalkMediator;
+using browser_sync::TalkMediatorImpl;
+using std::list;
+using std::hex;
+using std::string;
+using std::vector;
+using syncable::Directory;
+using syncable::DirectoryManager;
+
+static const int kServerReachablePollingIntervalMsec = 60000 * 60;
+static const int kThreadExitTimeoutMsec = 60000;
+static const int kSSLPort = 443;
+
+// We shouldn't call InitLogFiles more than once since that will cause a crash.
+// So we use a global state variable to avoid that. This doesn't work in case
+// of multiple threads, and if some other part also tries to call InitLogFiles
+// apart from this file. But this is okay for now since this is the only
+// place we call InitLogFiles.
+namespace {
+static bool g_log_files_initialized = false;
+static base::AtExitManager g_at_exit_manager;  // Necessary for NewCallback
+}  // empty namespace
+
+struct ThreadParams {
+  browser_sync::ServerConnectionManager* conn_mgr;
+#if defined(OS_WINDOWS)
+  HANDLE exit_flag;
+#endif
+};
+
+// This thread calls CheckServerReachable() whenever a change occurs
+// in the table that maps IP addresses to interfaces, for example when
+// the user unplugs his network cable.
+void* AddressWatchThread(void* arg) {
+  NameCurrentThreadForDebugging("SyncEngine_AddressWatcher");
+  LOG(INFO) << "starting the address watch thread";
+  const ThreadParams* const params = reinterpret_cast<const ThreadParams*>(arg);
+#if defined(OS_WINDOWS)
+  OVERLAPPED overlapped = {0};
+  overlapped.hEvent = CreateEvent(NULL, FALSE, TRUE, NULL);
+  HANDLE file;
+  DWORD rc = WAIT_OBJECT_0;
+  while (true) {
+    // Only call NotifyAddrChange() after the IP address has changed or if this
+    // is the first time through the loop.
+    if (WAIT_OBJECT_0 == rc) {
+      ResetEvent(overlapped.hEvent);
+      DWORD notify_result = NotifyAddrChange(&file, &overlapped);
+      if (ERROR_IO_PENDING != notify_result) {
+        LOG(ERROR) << "NotifyAddrChange() returned unexpected result "
+            << hex << notify_result;
+        break;
+      }
+    }
+    HANDLE events[] = { overlapped.hEvent, params->exit_flag };
+    rc = WaitForMultipleObjects(ARRAYSIZE(events), events, FALSE,
+                                kServerReachablePollingIntervalMsec);
+
+    // If the exit flag was signaled, the thread will exit.
+    if (WAIT_OBJECT_0 + 1 == rc)
+      break;
+
+    params->conn_mgr->CheckServerReachable();
+  }
+  CloseHandle(overlapped.hEvent);
+#else
+  // TODO(zork): Add this functionality to Linux.
+#endif
+  LOG(INFO) << "The address watch thread has stopped";
+  return 0;
+}
+
+namespace sync_api {
+class ModelSafeWorkerBridge;
+
+static const PSTR_CHAR kBookmarkSyncUserSettingsDatabase[] =
+    PSTR("BookmarkSyncSettings.sqlite3");
+static const PSTR_CHAR kDefaultNameForNewNodes[] = PSTR(" ");
+
+// The list of names which are reserved for use by the server.
+static const char16* kForbiddenServerNames[] =
+    { STRING16(""), STRING16("."), STRING16("..") };
+
+//////////////////////////////////////////////////////////////////////////
+// Static helper functions.
+
+// Helper function to look up the int64 metahandle of an object given the ID
+// string.
+static int64 IdToMetahandle(syncable::BaseTransaction* trans,
+                            const syncable::Id& id) {
+  syncable::Entry entry(trans, syncable::GET_BY_ID, id);
+  if (!entry.good())
+    return kInvalidId;
+  return entry.Get(syncable::META_HANDLE);
+}
+
+// Checks whether |name| is a server-illegal name followed by zero or more space
+// characters.  The three server-illegal names are the empty string, dot, and
+// dot-dot.  Very long names (>255 bytes in UTF-8 Normalization Form C) are
+// also illegal, but are not considered here.
+static bool IsNameServerIllegalAfterTrimming(const string16& name) {
+  size_t untrimmed_count = name.find_last_not_of(' ') + 1;
+  for (int i = 0; i < arraysize(kForbiddenServerNames); ++i) {
+    if (name.compare(0, untrimmed_count, kForbiddenServerNames[i]) == 0)
+      return true;
+  }
+  return false;
+}
+
+static bool EndsWithSpace(const string16& string) {
+  return !string.empty() && *string.rbegin() == ' ';
+}
+
+static inline void String16ToPathString(const sync_char16 *in,
+                                        PathString *out) {
+  string16 in_str(in);
+#if defined(OS_WINDOWS)
+  out->assign(in_str);
+#else
+  UTF16ToUTF8(in_str.c_str(), in_str.length(), out);
+#endif
+}
+
+static inline void PathStringToString16(const PathString& in, string16* out) {
+#if defined(OS_WINDOWS)
+  out->assign(in);
+#else
+  UTF8ToUTF16(in.c_str(), in.length(), out);
+#endif
+}
+
+// When taking a name from the syncapi, append a space if it matches the
+// pattern of a server-illegal name followed by zero or more spaces.
+static void SyncAPINameToServerName(const sync_char16 *sync_api_name,
+                                    PathString* out) {
+  String16ToPathString(sync_api_name, out);
+  string16 sync_api_name_str(sync_api_name);
+  if (IsNameServerIllegalAfterTrimming(sync_api_name_str))
+    out->append(PSTR(" "));
+}
+
+// In the reverse direction, if a server name matches the pattern of a
+// server-illegal name followed by one or more spaces, remove the trailing
+// space.
+static void ServerNameToSyncAPIName(const PathString& server_name,
+                                    string16*out) {
+  string16 server_name_str;
+  PathStringToString16(server_name, &server_name_str);
+  if (IsNameServerIllegalAfterTrimming(server_name_str) &&
+      EndsWithSpace(server_name_str))
+    out->assign(server_name_str, 0, server_name_str.size() - 1);
+  else
+    out->assign(server_name_str);
+}
+
+// A UserShare encapsulates the syncable pieces that represent an authenticated
+// user and their data (share).
+// This encompasses all pieces required to build transaction objects on the
+// syncable share.
+struct UserShare {
+  // The DirectoryManager itself, which is the parent of Transactions and can
+  // be shared across multiple threads (unlike Directory).
+  scoped_ptr<DirectoryManager> dir_manager;
+
+  // The username of the sync user. This is empty until we have performed at
+  // least one successful GAIA authentication with this username, which means
+  // on first-run it is empty until an AUTH_SUCCEEDED event and on future runs
+  // it is set as soon as the client instructs us to authenticate for the last
+  // known valid user (AuthenticateForLastKnownUser()).
+  // Stored as a PathString to avoid string conversions each time a transaction
+  // is created.
+  PathString authenticated_name;
+};
+
+////////////////////////////////////
+// BaseNode member definitions.
+
+// BaseNode::BaseNodeInternal provides storage for member Get() functions that
+// need to return pointers (e.g. strings).
+struct BaseNode::BaseNodeInternal {
+  string16 url;
+  string16 title;
+  Directory::ChildHandles child_handles;
+  syncable::Blob favicon;
+};
+
+BaseNode::BaseNode() : data_(new BaseNode::BaseNodeInternal) {}
+
+BaseNode::~BaseNode() {
+  delete data_;
+}
+
+int64 BaseNode::GetParentId() const {
+  return IdToMetahandle(GetTransaction()->GetWrappedTrans(),
+                        GetEntry()->Get(syncable::PARENT_ID));
+}
+
+int64 BaseNode::GetId() const {
+  return GetEntry()->Get(syncable::META_HANDLE);
+}
+
+bool BaseNode::GetIsFolder() const {
+  return GetEntry()->Get(syncable::IS_DIR);
+}
+
+const sync_char16* BaseNode::GetTitle() const {
+  // Store the string in data_ so that the returned pointer is valid.
+  ServerNameToSyncAPIName(GetEntry()->GetName().non_unique_value(),
+                         &data_->title);
+  return data_->title.c_str();
+}
+
+const sync_char16* BaseNode::GetURL() const {
+  // Store the string in data_ so that the returned pointer is valid.
+  PathStringToString16(GetEntry()->Get(syncable::BOOKMARK_URL), &data_->url);
+  return data_->url.c_str();
+}
+
+const int64* BaseNode::GetChildIds(size_t* child_count) const {
+  DCHECK(child_count);
+  Directory* dir = GetTransaction()->GetLookup();
+  dir->GetChildHandles(GetTransaction()->GetWrappedTrans(),
+                       GetEntry()->Get(syncable::ID), &data_->child_handles);
+
+  *child_count = data_->child_handles.size();
+  return (data_->child_handles.empty()) ? NULL : &data_->child_handles[0];
+}
+
+int64 BaseNode::GetPredecessorId() const {
+  syncable::Id id_string = GetEntry()->Get(syncable::PREV_ID);
+  if (id_string.IsRoot())
+    return kInvalidId;
+  return IdToMetahandle(GetTransaction()->GetWrappedTrans(), id_string);
+}
+
+int64 BaseNode::GetSuccessorId() const {
+  syncable::Id id_string = GetEntry()->Get(syncable::NEXT_ID);
+  if (id_string.IsRoot())
+    return kInvalidId;
+  return IdToMetahandle(GetTransaction()->GetWrappedTrans(), id_string);
+}
+
+int64 BaseNode::GetFirstChildId() const {
+  syncable::Directory* dir = GetTransaction()->GetLookup();
+  syncable::BaseTransaction* trans = GetTransaction()->GetWrappedTrans();
+  syncable::Id id_string =
+      dir->GetFirstChildId(trans, GetEntry()->Get(syncable::ID));
+  if (id_string.IsRoot())
+    return kInvalidId;
+  return IdToMetahandle(GetTransaction()->GetWrappedTrans(), id_string);
+}
+
+const unsigned char* BaseNode::GetFaviconBytes(size_t* size_in_bytes) {
+  data_->favicon = GetEntry()->Get(syncable::BOOKMARK_FAVICON);
+  *size_in_bytes = data_->favicon.size();
+  if (*size_in_bytes)
+    return &(data_->favicon[0]);
+  else
+    return NULL;
+}
+
+int64 BaseNode::GetExternalId() const {
+  return GetEntry()->Get(syncable::LOCAL_EXTERNAL_ID);
+}
+
+////////////////////////////////////
+// WriteNode member definitions
+void WriteNode::SetIsFolder(bool folder) {
+  if (entry_->Get(syncable::IS_DIR) == folder)
+    return;  // Skip redundant changes.
+
+  entry_->Put(syncable::IS_DIR, folder);
+  MarkForSyncing();
+}
+
+void WriteNode::SetTitle(const sync_char16* title) {
+  PathString server_legal_name;
+  SyncAPINameToServerName(title, &server_legal_name);
+  syncable::SyncName sync_name(server_legal_name);
+  syncable::DBName db_name(sync_name.value());
+  db_name.MakeOSLegal();
+  db_name.MakeNoncollidingForEntry(transaction_->GetWrappedTrans(),
+                                   entry_->Get(syncable::PARENT_ID), entry_);
+
+  syncable::Name new_name = syncable::Name::FromDBNameAndSyncName(db_name,
+                                                              sync_name);
+  if (new_name == entry_->GetName())
+    return;  // Skip redundant changes.
+
+  entry_->PutName(new_name);
+  MarkForSyncing();
+}
+
+void WriteNode::SetURL(const sync_char16* url) {
+  PathString url_string;
+  String16ToPathString(url, &url_string);
+  if (url_string == entry_->Get(syncable::BOOKMARK_URL))
+    return;  // Skip redundant changes.
+
+  entry_->Put(syncable::BOOKMARK_URL, url_string);
+  MarkForSyncing();
+}
+
+void WriteNode::SetExternalId(int64 id) {
+  if (GetExternalId() != id)
+    entry_->Put(syncable::LOCAL_EXTERNAL_ID, id);
+}
+
+WriteNode::WriteNode(WriteTransaction* transaction)
+    : entry_(NULL), transaction_(transaction) {
+  DCHECK(transaction);
+}
+
+WriteNode::~WriteNode() {
+  delete entry_;
+}
+
+// Find an existing node matching the ID |id|, and bind this WriteNode
+// to it.  Return true on success.
+bool WriteNode::InitByIdLookup(int64 id) {
+  DCHECK(!entry_) << "Init called twice";
+  DCHECK_NE(id, kInvalidId);
+  entry_ = new syncable::MutableEntry(transaction_->GetWrappedWriteTrans(),
+                                      syncable::GET_BY_HANDLE, id);
+  return (entry_->good() && !entry_->Get(syncable::IS_DEL));
+}
+
+// Create a new node with default properties, and bind this WriteNode to it.
+// Return true on success.
+bool WriteNode::InitByCreation(const BaseNode& parent,
+                               const BaseNode* predecessor) {
+  DCHECK(!entry_) << "Init called twice";
+  // |predecessor| must be a child of |parent| or NULL.
+  if (predecessor && predecessor->GetParentId() != parent.GetId()) {
+    DCHECK(false);
+    return false;
+  }
+
+  syncable::Id parent_id = parent.GetEntry()->Get(syncable::ID);
+
+  // Start out with a dummy name, but make it unique.  We expect
+  // the caller to set a meaningful name after creation.
+  syncable::DBName dummy(kDefaultNameForNewNodes);
+  dummy.MakeOSLegal();
+  dummy.MakeNoncollidingForEntry(transaction_->GetWrappedTrans(), parent_id,
+                                 NULL);
+
+  entry_ = new syncable::MutableEntry(transaction_->GetWrappedWriteTrans(),
+                                      syncable::CREATE, parent_id, dummy);
+
+  if (!entry_->good())
+    return false;
+
+  // Entries are untitled folders by default.
+  entry_->Put(syncable::IS_DIR, true);
+  // TODO(ncarter): Naming this bit IS_BOOKMARK_OBJECT is a bit unfortunate,
+  // since the rest of SyncAPI is essentially bookmark-agnostic.
+  entry_->Put(syncable::IS_BOOKMARK_OBJECT, true);
+
+  // Now set the predecessor, which sets IS_UNSYNCED as necessary.
+  PutPredecessor(predecessor);
+
+  return true;
+}
+
+bool WriteNode::SetPosition(const BaseNode& new_parent,
+                            const BaseNode* predecessor) {
+  // |predecessor| must be a child of |new_parent| or NULL.
+  if (predecessor && predecessor->GetParentId() != new_parent.GetId()) {
+    DCHECK(false);
+    return false;
+  }
+
+  syncable::Id new_parent_id = new_parent.GetEntry()->Get(syncable::ID);
+
+  // Filter out redundant changes if both the parent and the predecessor match.
+  if (new_parent_id == entry_->Get(syncable::PARENT_ID)) {
+    const syncable::Id& old = entry_->Get(syncable::PREV_ID);
+    if ((!predecessor && old.IsRoot()) ||
+        (predecessor && (old == predecessor->GetEntry()->Get(syncable::ID)))) {
+      return true;
+    }
+  }
+
+  // Discard the old database name, derive a new database name from the sync
+  // name, and make it legal and unique.
+  syncable::Name name = syncable::Name::FromSyncName(GetEntry()->GetName());
+  name.db_value().MakeOSLegal();
+  name.db_value().MakeNoncollidingForEntry(GetTransaction()->GetWrappedTrans(),
+                                           new_parent_id, entry_);
+
+  // Atomically change the parent and name. This will fail if it would
+  // introduce a cycle in the hierarchy.
+  if (!entry_->PutParentIdAndName(new_parent_id, name))
+    return false;
+
+  // Now set the predecessor, which sets IS_UNSYNCED as necessary.
+  PutPredecessor(predecessor);
+
+  return true;
+}
+
+const syncable::Entry* WriteNode::GetEntry() const {
+  return entry_;
+}
+
+const BaseTransaction* WriteNode::GetTransaction() const {
+  return transaction_;
+}
+
+void WriteNode::Remove() {
+  entry_->Put(syncable::IS_DEL, true);
+  MarkForSyncing();
+}
+
+void WriteNode::PutPredecessor(const BaseNode* predecessor) {
+  syncable::Id predecessor_id = predecessor ?
+      predecessor->GetEntry()->Get(syncable::ID) : syncable::Id();
+  entry_->PutPredecessor(predecessor_id);
+  // Mark this entry as unsynced, to wake up the syncer.
+  MarkForSyncing();
+}
+
+void WriteNode::SetFaviconBytes(const unsigned char* bytes,
+                                size_t size_in_bytes) {
+  syncable::Blob new_favicon(bytes, bytes + size_in_bytes);
+  if (new_favicon == entry_->Get(syncable::BOOKMARK_FAVICON))
+    return;  // Skip redundant changes.
+
+  entry_->Put(syncable::BOOKMARK_FAVICON, new_favicon);
+  MarkForSyncing();
+}
+
+void WriteNode::MarkForSyncing() {
+  syncable::MarkForSyncing(entry_);
+}
+
+//////////////////////////////////////////////////////////////////////////
+// ReadNode member definitions
+ReadNode::ReadNode(const BaseTransaction* transaction)
+    : entry_(NULL), transaction_(transaction) {
+  DCHECK(transaction);
+}
+
+ReadNode::~ReadNode() {
+  delete entry_;
+}
+
+void ReadNode::InitByRootLookup() {
+  DCHECK(!entry_) << "Init called twice";
+  syncable::BaseTransaction* trans = transaction_->GetWrappedTrans();
+  entry_ = new syncable::Entry(trans, syncable::GET_BY_ID, trans->root_id());
+  if (!entry_->good())
+    DCHECK(false) << "Could not lookup root node for reading.";
+}
+
+bool ReadNode::InitByIdLookup(int64 id) {
+  DCHECK(!entry_) << "Init called twice";
+  DCHECK_NE(id, kInvalidId);
+  syncable::BaseTransaction* trans = transaction_->GetWrappedTrans();
+  entry_ = new syncable::Entry(trans, syncable::GET_BY_HANDLE, id);
+  if (!entry_->good())
+    return false;
+  if (entry_->Get(syncable::IS_DEL))
+    return false;
+  LOG_IF(WARNING, !entry_->Get(syncable::IS_BOOKMARK_OBJECT))
+      << "SyncAPI InitByIdLookup referencing non-bookmark object.";
+  return true;
+}
+
+const syncable::Entry* ReadNode::GetEntry() const {
+  return entry_;
+}
+
+const BaseTransaction* ReadNode::GetTransaction() const {
+  return transaction_;
+}
+
+bool ReadNode::InitByTagLookup(const sync_char16* tag) {
+  DCHECK(!entry_) << "Init called twice";
+  PathString tag_string;
+  String16ToPathString(tag, &tag_string);
+  if (tag_string.empty())
+    return false;
+  syncable::BaseTransaction* trans = transaction_->GetWrappedTrans();
+  entry_ = new syncable::Entry(trans, syncable::GET_BY_TAG, tag_string);
+  if (!entry_->good())
+    return false;
+  if (entry_->Get(syncable::IS_DEL))
+    return false;
+  LOG_IF(WARNING, !entry_->Get(syncable::IS_BOOKMARK_OBJECT))
+      << "SyncAPI InitByTagLookup referencing non-bookmark object.";
+  return true;
+}
+
+
+//////////////////////////////////////////////////////////////////////////
+// ReadTransaction member definitions
+ReadTransaction::ReadTransaction(UserShare* share)
+    : BaseTransaction(share),
+      transaction_(NULL) {
+  transaction_ = new syncable::ReadTransaction(GetLookup(), __FILE__, __LINE__);
+}
+
+ReadTransaction::~ReadTransaction() {
+  delete transaction_;
+}
+
+syncable::BaseTransaction* ReadTransaction::GetWrappedTrans() const {
+  return transaction_;
+}
+
+//////////////////////////////////////////////////////////////////////////
+// WriteTransaction member definitions
+WriteTransaction::WriteTransaction(UserShare* share)
+    : BaseTransaction(share),
+      transaction_(NULL) {
+  transaction_ = new syncable::WriteTransaction(GetLookup(), syncable::SYNCAPI,
+                                                __FILE__, __LINE__);
+}
+
+WriteTransaction::~WriteTransaction() {
+  delete transaction_;
+}
+
+syncable::BaseTransaction* WriteTransaction::GetWrappedTrans() const {
+  return transaction_;
+}
+
+// An implementation of Visitor that we use to "visit" the
+// ModelSafeWorkerInterface provided by a client of this API. The object we
+// visit is responsible for calling DoWork, which will invoke Run() on it's
+// cached work closure.
+class ModelSafeWorkerVisitor : public ModelSafeWorkerInterface::Visitor {
+ public:
+  explicit ModelSafeWorkerVisitor(Closure* work) : work_(work) { }
+  virtual ~ModelSafeWorkerVisitor() { }
+
+  // ModelSafeWorkerInterface::Visitor implementation.
+  virtual void DoWork() {
+    work_->Run();
+  }
+
+ private:
+  // The work to be done. We run this on DoWork and it cleans itself up
+  // after it is run.
+  Closure* work_;
+
+  DISALLOW_COPY_AND_ASSIGN(ModelSafeWorkerVisitor);
+};
+
+// This class is declared in the cc file to allow inheritance from sync types.
+// The ModelSafeWorkerBridge is a liason between a syncapi-client defined
+// ModelSafeWorkerInterface and the actual ModelSafeWorker used by the Syncer
+// for the current SyncManager.
+class ModelSafeWorkerBridge : public browser_sync::ModelSafeWorker {
+ public:
+  // Takes ownership of |worker|.
+  explicit ModelSafeWorkerBridge(ModelSafeWorkerInterface* worker)
+      : worker_(worker) {
+  }
+  virtual ~ModelSafeWorkerBridge() { }
+
+  // Overriding ModelSafeWorker.
+  virtual void DoWorkAndWaitUntilDone(Closure* work) {
+    // When the syncer has work to be done, we forward it to our worker who
+    // will invoke DoWork on |visitor| when appropriate (from model safe
+    // thread).
+    ModelSafeWorkerVisitor visitor(work);
+    worker_->CallDoWorkFromModelSafeThreadAndWait(&visitor);
+  }
+
+ private:
+  // The worker that we can forward work requests to, to ensure the work
+  // is performed on an appropriate model safe thread.
+  scoped_ptr<ModelSafeWorkerInterface> worker_;
+
+  DISALLOW_COPY_AND_ASSIGN(ModelSafeWorkerBridge);
+};
+
+// A GaiaAuthenticator that uses HttpPostProviders instead of CURL.
+class BridgedGaiaAuthenticator : public browser_sync::GaiaAuthenticator {
+ public:
+  BridgedGaiaAuthenticator(const string& user_agent, const string& service_id,
+                           const string& gaia_url,
+                           HttpPostProviderFactory* factory)
+      : GaiaAuthenticator(user_agent, service_id, gaia_url),
+        gaia_source_(user_agent), post_factory_(factory) {
+  }
+
+  virtual ~BridgedGaiaAuthenticator() {
+  }
+
+  virtual bool Post(const GURL& url, const string& post_body,
+                    unsigned long* response_code, string* response_body) {
+    string connection_url = "https://";
+    connection_url += url.host();
+    connection_url += url.path();
+    HttpPostProviderInterface* http = post_factory_->Create();
+    http->SetUserAgent(gaia_source_.c_str());
+    // SSL is on 443 for Gaia Posts always.
+    http->SetURL(connection_url.c_str(), kSSLPort);
+    http->SetPostPayload("application/x-www-form-urlencoded",
+                         post_body.length(), post_body.c_str());
+
+    int os_error_code = 0;
+    int int_response_code = 0;
+    if (!http->MakeSynchronousPost(&os_error_code, &int_response_code)) {
+      LOG(INFO) << "Http POST failed, error returns: " << os_error_code;
+      return false;
+    }
+    *response_code = static_cast<int>(int_response_code);
+    response_body->assign(http->GetResponseContent(),
+                          http->GetResponseContentLength());
+    post_factory_->Destroy(http);
+    return true;
+  }
+ private:
+  const std::string gaia_source_;
+  scoped_ptr<HttpPostProviderFactory> post_factory_;
+  DISALLOW_COPY_AND_ASSIGN(BridgedGaiaAuthenticator);
+};
+
+//////////////////////////////////////////////////////////////////////////
+// SyncManager's implementation: SyncManager::SyncInternal
+class SyncManager::SyncInternal {
+ public:
+  typedef PThreadScopedLock<PThreadMutex> MutexLock;
+  explicit SyncInternal(SyncManager* sync_manager)
+      : observer_(NULL),
+        command_channel_(0),
+        auth_problem_(AUTH_PROBLEM_NONE),
+        sync_manager_(sync_manager),
+        notification_pending_(false),
+        initialized_(false) {
+  }
+
+  ~SyncInternal() { }
+
+  bool Init(const PathString& database_location,
+            const std::string& sync_server_and_path,
+            int port,
+            const char* gaia_service_id,
+            const char* gaia_source,
+            bool use_ssl,
+            HttpPostProviderFactory* post_factory,
+            HttpPostProviderFactory* auth_post_factory,
+            ModelSafeWorkerInterface* model_safe_worker,
+            bool attempt_last_user_authentication,
+            const char* user_agent);
+
+  // Tell sync engine to submit credentials to GAIA for verification and start
+  // the syncing process on success. Successful GAIA authentication will kick
+  // off the following chain of events:
+  // 1. Cause sync engine to open the syncer database.
+  // 2. Trigger the AuthWatcher to create a Syncer for the directory and call
+  //    SyncerThread::SyncDirectory; the SyncerThread will block until (4).
+  // 3. Tell the ServerConnectionManager to pass the newly received GAIA auth
+  //    token to a sync server to obtain a sync token.
+  // 4. On receipt of this token, the ServerConnectionManager broadcasts
+  //    a server-reachable event, which will unblock the SyncerThread,
+  //    and the rest is the future.
+  //
+  // If authentication fails, an event will be broadcast all the way up to
+  // the SyncManager::Observer. It may, in turn, decide to try again with new
+  // credentials. Calling this method again is the appropriate course of action
+  // to "retry".
+  void Authenticate(const std::string& username, const std::string& password);
+
+  // Call periodically from a database-safe thread to persist recent changes
+  // to the syncapi model.
+  void SaveChanges();
+
+  // This listener is called upon completion of a syncable transaction, and
+  // builds the list of sync-engine initiated changes that will be forwarded to
+  // the SyncManager's Observers.
+  void HandleChangeEvent(const syncable::DirectoryChangeEvent& event);
+  void HandleTransactionCompleteChangeEvent(
+      const syncable::DirectoryChangeEvent& event);
+  void HandleCalculateChangesChangeEventFromSyncApi(
+      const syncable::DirectoryChangeEvent& event);
+  void HandleCalculateChangesChangeEventFromSyncer(
+      const syncable::DirectoryChangeEvent& event);
+
+  // This listener is called by the syncer channel for all syncer events.
+  void HandleSyncerEvent(const SyncerEvent& event);
+
+  // We have a direct hookup to the authwatcher to be notified for auth failures
+  // on startup, to serve our UI needs.
+  void HandleAuthWatcherEvent(const AuthWatcherEvent& event);
+
+  // Accessors for the private members.
+  DirectoryManager* dir_manager() { return share_.dir_manager.get(); }
+  SyncAPIServerConnectionManager* connection_manager() {
+    return connection_manager_.get();
+  }
+  SyncerThread* syncer_thread() { return syncer_thread_.get(); }
+  TalkMediator* talk_mediator() { return talk_mediator_.get(); }
+  AuthWatcher* auth_watcher() { return auth_watcher_.get(); }
+  AllStatus* allstatus() { return &allstatus_; }
+  void set_observer(Observer* observer) { observer_ = observer; }
+  UserShare* GetUserShare() { return &share_; }
+
+  // Return the currently active (validated) username as a PathString for
+  // use with syncable types.
+  const PathString& username_for_share() const {
+    return share_.authenticated_name;
+  }
+
+  // Returns the authenticated username from our AuthWatcher in UTF8.
+  // See SyncManager::GetAuthenticatedUsername for details.
+  const char* GetAuthenticatedUsername();
+
+  // Note about SyncManager::Status implementation: Status is a trimmed
+  // down AllStatus::Status, augmented with authentication failure information
+  // gathered from the internal AuthWatcher. The sync UI itself hooks up to
+  // various sources like the AuthWatcher individually, but with syncapi we try
+  // to keep everything status-related in one place. This means we have to
+  // privately manage state about authentication failures, and whenever the
+  // status or status summary is requested we aggregate this state with
+  // AllStatus::Status information.
+  Status ComputeAggregatedStatus();
+  Status::Summary ComputeAggregatedStatusSummary();
+
+  // See SyncManager::SetupForTestMode for information.
+  void SetupForTestMode(const sync_char16* test_username);
+
+  // See SyncManager::Shutdown for information.
+  void Shutdown();
+
+  // Whether we're initialized to the point of being able to accept changes
+  // (and hence allow transaction creation). See initialized_ for details.
+  bool initialized() const {
+    MutexLock lock(&initialized_mutex_);
+    return initialized_;
+  }
+ private:
+  // Try to authenticate using persisted credentials from a previous successful
+  // authentication. If no such credentials exist, calls OnAuthError on
+  // the client to collect credentials. Otherwise, there exist local
+  // credentials that were once used for a successful auth, so we'll try to
+  // re-use these.
+  // Failure of that attempt will be communicated as normal using
+  // OnAuthError. Since this entry point will bypass normal GAIA
+  // authentication and try to authenticate directly with the sync service
+  // using a cached token, authentication failure will generally occur due to
+  // expired credentials, or possibly because of a password change.
+  void AuthenticateForLastKnownUser();
+
+  // Helper to call OnAuthError when no authentication credentials
+  // are available.
+  void RaiseAuthNeededEvent();
+
+  // Helper to set initialized_ to true and raise an event to clients to
+  // notify that initialization is complete and it is safe to send us changes.
+  // If already initialized, this is a no-op.
+  void MarkAndNotifyInitializationComplete();
+
+  // Determine if the parents or predecessors differ between the old and new
+  // versions of an entry stored in |a| and |b|.  Note that a node's index
+  // may change without its NEXT_ID changing if the node at NEXT_ID also
+  // moved (but the relative order is unchanged).  To handle such cases,
+  // we rely on the caller to treat a position update on any sibling as
+  // updating the positions of all siblings.
+  static bool BookmarkPositionsDiffer(const syncable::EntryKernel& a,
+                                      const syncable::Entry& b) {
+    if (a.ref(syncable::NEXT_ID) != b.Get(syncable::NEXT_ID))
+      return true;
+    if (a.ref(syncable::PARENT_ID) != b.Get(syncable::PARENT_ID))
+      return true;
+    return false;
+  }
+
+  // Determine if any of the fields made visible to clients of the Sync API
+  // differ between the versions of an entry stored in |a| and |b|.
+  // A return value of false means that it should be OK to ignore this change.
+  static bool BookmarkPropertiesDiffer(const syncable::EntryKernel& a,
+                                       const syncable::Entry& b) {
+    if (a.ref(syncable::NAME) != b.Get(syncable::NAME))
+      return true;
+    if (a.ref(syncable::UNSANITIZED_NAME) != b.Get(syncable::UNSANITIZED_NAME))
+      return true;
+    if (a.ref(syncable::IS_DIR) != b.Get(syncable::IS_DIR))
+      return true;
+    if (a.ref(syncable::BOOKMARK_URL) != b.Get(syncable::BOOKMARK_URL))
+      return true;
+    if (a.ref(syncable::BOOKMARK_FAVICON) != b.Get(syncable::BOOKMARK_FAVICON))
+      return true;
+    if (BookmarkPositionsDiffer(a, b))
+      return true;
+    return false;
+  }
+
+  // We couple the DirectoryManager and username together in a UserShare member
+  // so we can return a handle to share_ to clients of the API for use when
+  // constructing any transaction type.
+  UserShare share_;
+
+  // A cached string for callers of GetAuthenticatedUsername. We just store the
+  // last result of auth_watcher_->email() here and change it on future calls,
+  // because callers of GetAuthenticatedUsername are supposed to copy the value
+  // if they need it for longer than the scope of the call.
+  std::string cached_auth_watcher_email_;
+
+  // A wrapper around a sqlite store used for caching authentication data,
+  // last user information, current sync-related URLs, and more.
+  scoped_ptr<UserSettings> user_settings_;
+
+  // Observer registered via SetObserver/RemoveObserver.
+  // WARNING: This can be NULL!
+  Observer* observer_;
+
+  // A sink for client commands from the syncer needed to create a SyncerThread.
+  ClientCommandChannel command_channel_;
+
+  // The ServerConnectionManager used to abstract communication between
+  // the client (the Syncer) and the sync server.
+  scoped_ptr<SyncAPIServerConnectionManager> connection_manager_;
+
+  // The thread that runs the Syncer. Needs to be explicitly Start()ed.
+  scoped_ptr<SyncerThread> syncer_thread_;
+
+  // Notification (xmpp) handler.
+  scoped_ptr<TalkMediator> talk_mediator_;
+
+  // A multi-purpose status watch object that aggregates stats from various
+  // sync components.
+  AllStatus allstatus_;
+
+  // AuthWatcher kicks off the authentication process and follows it through
+  // phase 1 (GAIA) to phase 2 (sync engine). As part of this work it determines
+  // the initial connectivity and causes the server connection event to be
+  // broadcast, which signals the syncer thread to start syncing.
+  // It has a heavy duty constructor requiring boilerplate so we heap allocate.
+  scoped_ptr<AuthWatcher> auth_watcher_;
+
+  // A store of change records produced by HandleChangeEvent during the
+  // CALCULATE_CHANGES step, and to be processed, and forwarded to the
+  // observer, by HandleChangeEvent during the TRANSACTION_COMPLETE step.
+  ChangeReorderBuffer change_buffer_;
+
+  // The event listener hookup that is registered for HandleChangeEvent.
+  scoped_ptr<EventListenerHookup> dir_change_hookup_;
+
+  // The event listener hookup registered for HandleSyncerEvent.
+  scoped_ptr<EventListenerHookup> syncer_event_;
+
+  // The event listener hookup registered for HandleAuthWatcherEvent.
+  scoped_ptr<EventListenerHookup> authwatcher_hookup_;
+
+  // Our cache of a recent authentication problem. If no authentication problem
+  // occurred, or if the last problem encountered has been cleared (by a
+  // subsequent AuthWatcherEvent), this is set to AUTH_PROBLEM_NONE.
+  AuthProblem auth_problem_;
+
+  // The sync dir_manager to which we belong.
+  SyncManager* const sync_manager_;
+
+  // Parameters for our thread listening to network status changes.
+  ThreadParams address_watch_params_;
+  thread_handle address_watch_thread_;
+
+  // True if the next SyncCycle should notify peers of an update.
+  bool notification_pending_;
+
+  // Set to true once Init has been called, and we know of an authenticated
+  // valid) username either from a fresh authentication attempt (as in
+  // first-use case) or from a previous attempt stored in our UserSettings
+  // (as in the steady-state), and the syncable::Directory has been opened,
+  // meaning we are ready to accept changes.  Protected by initialized_mutex_
+  // as it can get read/set by both the SyncerThread and the AuthWatcherThread.
+  bool initialized_;
+  mutable PThreadMutex initialized_mutex_;
+};
+
+SyncManager::SyncManager() {
+  data_ = new SyncInternal(this);
+}
+
+bool SyncManager::Init(const sync_char16* database_location,
+                       const char* sync_server_and_path,
+                       int sync_server_port,
+                       const char* gaia_service_id,
+                       const char* gaia_source,
+                       bool use_ssl,
+                       HttpPostProviderFactory* post_factory,
+                       HttpPostProviderFactory* auth_post_factory,
+                       ModelSafeWorkerInterface* model_safe_worker,
+                       bool attempt_last_user_authentication,
+                       const char* user_agent) {
+  DCHECK(database_location);
+  DCHECK(post_factory);
+
+  PathString db_path;
+  String16ToPathString(database_location, &db_path);
+  string server_string(sync_server_and_path);
+  return data_->Init(db_path,
+                     server_string,
+                     sync_server_port,
+                     gaia_service_id,
+                     gaia_source,
+                     use_ssl,
+                     post_factory,
+                     auth_post_factory,
+                     model_safe_worker,
+                     attempt_last_user_authentication,
+                     user_agent);
+}
+
+void SyncManager::Authenticate(const char* username, const char* password) {
+  data_->Authenticate(std::string(username), std::string(password));
+}
+
+const char* SyncManager::GetAuthenticatedUsername() {
+  if (!data_)
+    return NULL;
+  return data_->GetAuthenticatedUsername();
+}
+
+const char* SyncManager::SyncInternal::GetAuthenticatedUsername() {
+  cached_auth_watcher_email_ = browser_sync::ToUTF8(
+      username_for_share()).get_string();
+  return cached_auth_watcher_email_.c_str();
+}
+
+bool SyncManager::SyncInternal::Init(
+    const PathString& database_location,
+    const std::string& sync_server_and_path,
+    int port,
+    const char* gaia_service_id,
+    const char* gaia_source,
+    bool use_ssl, HttpPostProviderFactory* post_factory,
+    HttpPostProviderFactory* auth_post_factory,
+    ModelSafeWorkerInterface* model_safe_worker,
+    bool attempt_last_user_authentication,
+    const char* user_agent) {
+
+  if (!g_log_files_initialized) {
+    // TODO(timsteele): Call InitLogFiles() or equivalent.
+    g_log_files_initialized = true;
+  }
+
+  // Set up UserSettings, creating the db if necessary. We need this to
+  // instantiate a URLFactory to give to the Syncer.
+  PathString settings_db_file = AppendSlash(database_location) +
+                                kBookmarkSyncUserSettingsDatabase;
+  user_settings_.reset(new UserSettings());
+  if (!user_settings_->Init(settings_db_file))
+    return false;
+
+  share_.dir_manager.reset(new DirectoryManager(database_location));
+
+  string client_id = user_settings_->GetClientId();
+  connection_manager_.reset(new SyncAPIServerConnectionManager(
+      sync_server_and_path, port, use_ssl, user_agent, client_id));
+
+  // TODO(timsteele): This is temporary windows crap needed to listen for
+  // network status changes. We should either pump this up to the embedder to
+  // do (and call us in CheckServerReachable, for ex), or at least make this
+  // platform independent in here.
+  // TODO(ncarter): When this gets cleaned up, the implementation of
+  // CreatePThread can also be removed.
+#if defined(OS_WINDOWS)
+  HANDLE exit_flag = CreateEvent(NULL, TRUE /*manual reset*/, FALSE, NULL);
+  address_watch_params_.exit_flag = exit_flag;
+#endif
+  address_watch_params_.conn_mgr = connection_manager();
+  address_watch_thread_ = CreatePThread(AddressWatchThread,
+                                        &address_watch_params_);
+  DCHECK(NULL != address_watch_thread_);
+
+  // Hand over the bridged POST factory to be owned by the connection
+  // dir_manager.
+  connection_manager()->SetHttpPostProviderFactory(post_factory);
+
+  // Watch various objects for aggregated status.
+  allstatus()->WatchConnectionManager(connection_manager());
+
+  std::string gaia_url = browser_sync::kGaiaUrl;
+  const char* service_id = gaia_service_id ?
+      gaia_service_id : SYNC_SERVICE_NAME;
+
+  talk_mediator_.reset(new TalkMediatorImpl());
+  allstatus()->WatchTalkMediator(talk_mediator());
+
+  BridgedGaiaAuthenticator* gaia_auth = new BridgedGaiaAuthenticator(
+      gaia_source, service_id, gaia_url, auth_post_factory);
+
+  auth_watcher_.reset(new AuthWatcher(dir_manager(),
+                                      connection_manager(),
+                                      &allstatus_,
+                                      gaia_source,
+                                      service_id,
+                                      gaia_url,
+                                      user_settings_.get(),
+                                      gaia_auth,
+                                      talk_mediator()));
+
+  talk_mediator()->WatchAuthWatcher(auth_watcher());
+  allstatus()->WatchAuthWatcher(auth_watcher());
+  authwatcher_hookup_.reset(NewEventListenerHookup(auth_watcher_->channel(),
+      this, &SyncInternal::HandleAuthWatcherEvent));
+
+  // Tell the SyncerThread to use the ModelSafeWorker for bookmark model work.
+  // We set up both sides of the "bridge" here, with the ModelSafeWorkerBridge
+  // on the Syncer side, and |model_safe_worker| on the API client side.
+  ModelSafeWorkerBridge* worker = new ModelSafeWorkerBridge(model_safe_worker);
+
+  syncer_thread_.reset(new SyncerThread(&command_channel_,
+                                        dir_manager(),
+                                        connection_manager(),
+                                        &allstatus_,
+                                        worker));
+  syncer_thread()->WatchTalkMediator(talk_mediator());
+  allstatus()->WatchSyncerThread(syncer_thread());
+
+  syncer_thread()->Start();  // Start the syncer thread. This won't actually
+                             // result in any syncing until at least the
+                             // DirectoryManager broadcasts the OPENED event,
+                             // and a valid server connection is detected.
+
+  if (attempt_last_user_authentication)
+    AuthenticateForLastKnownUser();
+  return true;
+}
+
+void SyncManager::SyncInternal::MarkAndNotifyInitializationComplete() {
+  // There is only one real time we need this mutex.  If we get an auth
+  // success, and before the initial sync ends we get an auth failure.  In this
+  // case we'll be listening to both the AuthWatcher and Syncer, and it's a race
+  // between their respective threads to call MarkAndNotify.  We need to make
+  // sure the observer is notified once and only once.
+  {
+    MutexLock lock(&initialized_mutex_);
+    if (initialized_)
+      return;
+    initialized_ = true;
+  }
+
+  // Notify that initialization is complete.
+  if (observer_)
+    observer_->OnInitializationComplete();
+}
+
+void SyncManager::SyncInternal::Authenticate(const std::string& username,
+                                             const std::string& password) {
+  DCHECK(username_for_share().empty() ||
+        (username == browser_sync::ToUTF8(username_for_share()).get_string()))
+        << "Username change from valid username detected";
+  if (allstatus()->status().authenticated)
+    return;
+  if (password.empty()) {
+    // TODO(timsteele): Seems like this shouldn't be needed, but auth_watcher
+    // currently drops blank password attempts on the floor and doesn't update
+    // state; it only LOGs an error in this case. We want to make sure we set
+    // our AuthProblem state to denote an error.
+    RaiseAuthNeededEvent();
+  }
+  auth_watcher()->Authenticate(username, password, true);
+}
+
+void SyncManager::SyncInternal::AuthenticateForLastKnownUser() {
+  std::string username;
+  std::string auth_token;
+  if (!(auth_watcher()->settings()->GetLastUserAndServiceToken(
+        SYNC_SERVICE_NAME, &username, &auth_token))) {
+    RaiseAuthNeededEvent();
+    return;
+  }
+
+  browser_sync::ToPathString s(username);
+  if (s.good()) {
+    share_.authenticated_name = s.get_string16();
+  } else {
+    RaiseAuthNeededEvent();
+    return;
+  }
+
+  // We optimize by opening the directory before the "fresh" authentication
+  // attempt completes so that we can immediately begin processing changes.
+  if (!dir_manager()->Open(username_for_share())) {
+    DCHECK(false) << "Had last known user but could not open directory";
+    return;
+  }
+
+  // Set the sync data type so that the server only sends us bookmarks
+  // changes.
+  {
+    syncable::ScopedDirLookup lookup(dir_manager(), username_for_share());
+    if (!lookup.good()) {
+      DCHECK(false) << "ScopedDirLookup failed on successfully opened dir";
+      return;
+    }
+    if (lookup->initial_sync_ended())
+      MarkAndNotifyInitializationComplete();
+  }
+
+  auth_watcher()->AuthenticateWithToken(username, auth_token);
+}
+
+void SyncManager::SyncInternal::RaiseAuthNeededEvent() {
+  auth_problem_ = AUTH_PROBLEM_INVALID_GAIA_CREDENTIALS;
+  if (observer_)
+    observer_->OnAuthProblem(auth_problem_);
+}
+
+SyncManager::~SyncManager() {
+  delete data_;
+}
+
+void SyncManager::SetObserver(Observer* observer) {
+  data_->set_observer(observer);
+}
+
+void SyncManager::RemoveObserver() {
+  data_->set_observer(NULL);
+}
+
+void SyncManager::Shutdown() {
+  data_->Shutdown();
+}
+
+void SyncManager::SyncInternal::Shutdown() {
+  // First reset the AuthWatcher in case an auth attempt is in progress so that
+  // it terminates gracefully before we shutdown and close other components.
+  // Otherwise the attempt can complete after we've closed the directory, for
+  // example, and cause initialization to continue, which is bad.
+  auth_watcher_.reset();
+
+  if (syncer_thread()) {
+    if (!syncer_thread()->Stop(kThreadExitTimeoutMsec))
+      DCHECK(false) << "Unable to stop the syncer, it won't be happy...";
+  }
+
+  // Shutdown the xmpp buzz connection.
+  LOG(INFO) << "P2P: Mediator logout started.";
+  if (talk_mediator()) {
+    talk_mediator()->Logout();
+  }
+  LOG(INFO) << "P2P: Mediator logout completed.";
+
+  if (dir_manager()) {
+    dir_manager()->FinalSaveChangesForAll();
+    dir_manager()->Close(username_for_share());
+  }
+
+  // Reset the DirectoryManager and UserSettings so they relinquish sqlite
+  // handles to backing files.
+  share_.dir_manager.reset();
+  user_settings_.reset();
+
+  // We don't want to process any more events.
+  dir_change_hookup_.reset();
+  syncer_event_.reset();
+  authwatcher_hookup_.reset();
+
+#if defined(OS_WINDOWS)
+  // Stop the address watch thread by signaling the exit flag.
+  // TODO(timsteele): Same as todo in Init().
+  SetEvent(address_watch_params_.exit_flag);
+  const DWORD wait_result = WaitForSingleObject(address_watch_thread_,
+                                                kThreadExitTimeoutMsec);
+  LOG_IF(ERROR, WAIT_FAILED == wait_result) << "Waiting for addr change thread "
+      "to exit failed. GetLastError(): " << hex << GetLastError();
+  LOG_IF(ERROR, WAIT_TIMEOUT == wait_result) << "Thread exit timeout expired";
+  CloseHandle(address_watch_params_.exit_flag);
+#endif
+}
+
+// Listen to model changes, filter out ones initiated by the sync API, and
+// saves the rest (hopefully just backend Syncer changes resulting from
+// ApplyUpdates) to data_->changelist.
+void SyncManager::SyncInternal::HandleChangeEvent(
+    const syncable::DirectoryChangeEvent& event) {
+  if (event.todo == syncable::DirectoryChangeEvent::TRANSACTION_COMPLETE) {
+    HandleTransactionCompleteChangeEvent(event);
+    return;
+  } else if (event.todo == syncable::DirectoryChangeEvent::CALCULATE_CHANGES) {
+    if (event.writer == syncable::SYNCAPI) {
+      HandleCalculateChangesChangeEventFromSyncApi(event);
+      return;
+    }
+    HandleCalculateChangesChangeEventFromSyncer(event);
+    return;
+  } else if (event.todo == syncable::DirectoryChangeEvent::SHUTDOWN) {
+    dir_change_hookup_.reset();
+  }
+}
+
+void SyncManager::SyncInternal::HandleTransactionCompleteChangeEvent(
+    const syncable::DirectoryChangeEvent& event) {
+  DCHECK_EQ(event.todo, syncable::DirectoryChangeEvent::TRANSACTION_COMPLETE);
+  // This notification happens immediately after a syncable WriteTransaction
+  // falls out of scope.
+  if (change_buffer_.IsEmpty() || !observer_)
+    return;
+
+  ReadTransaction trans(GetUserShare());
+  vector<ChangeRecord> ordered_changes;
+  change_buffer_.GetAllChangesInTreeOrder(&trans, &ordered_changes);
+  if (!ordered_changes.empty()) {
+    observer_->OnChangesApplied(&trans, &ordered_changes[0],
+                                ordered_changes.size());
+  }
+  change_buffer_.Clear();
+}
+
+void SyncManager::SyncInternal::HandleCalculateChangesChangeEventFromSyncApi(
+    const syncable::DirectoryChangeEvent& event) {
+  // We have been notified about a user action changing the bookmark model.
+  DCHECK_EQ(event.todo, syncable::DirectoryChangeEvent::CALCULATE_CHANGES);
+  DCHECK_EQ(event.writer, syncable::SYNCAPI);
+  LOG_IF(WARNING, !change_buffer_.IsEmpty()) <<
+      "CALCULATE_CHANGES called with unapplied old changes.";
+
+  bool exists_unsynced_items = false;
+  for (syncable::OriginalEntries::const_iterator i = event.originals->begin();
+       i != event.originals->end() && !exists_unsynced_items;
+       ++i) {
+    int64 id = i->ref(syncable::META_HANDLE);
+    syncable::Entry e(event.trans, syncable::GET_BY_HANDLE, id);
+    DCHECK(e.good());
+
+    if (e.IsRoot()) {
+      // Ignore root object, should it ever change.
+      continue;
+    } else if (!e.Get(syncable::IS_BOOKMARK_OBJECT)) {
+      // Ignore non-bookmark objects.
+      continue;
+    } else if (e.Get(syncable::IS_UNSYNCED)) {
+      // Unsynced items will cause us to nudge the the syncer.
+      exists_unsynced_items = true;
+    }
+  }
+  if (exists_unsynced_items && syncer_thread()) {
+    syncer_thread()->NudgeSyncer(200, SyncerThread::kLocal);  // 1/5 a second.
+  }
+}
+
+void SyncManager::SyncInternal::HandleCalculateChangesChangeEventFromSyncer(
+    const syncable::DirectoryChangeEvent& event) {
+  // We only expect one notification per sync step, so change_buffer_ should
+  // contain no pending entries.
+  DCHECK_EQ(event.todo, syncable::DirectoryChangeEvent::CALCULATE_CHANGES);
+  DCHECK_EQ(event.writer, syncable::SYNCER);
+  LOG_IF(WARNING, !change_buffer_.IsEmpty()) <<
+      "CALCULATE_CHANGES called with unapplied old changes.";
+
+  for (syncable::OriginalEntries::const_iterator i = event.originals->begin();
+       i != event.originals->end(); ++i) {
+    int64 id = i->ref(syncable::META_HANDLE);
+    syncable::Entry e(event.trans, syncable::GET_BY_HANDLE, id);
+    bool existed_before = !i->ref(syncable::IS_DEL);
+    bool exists_now = e.good() && !e.Get(syncable::IS_DEL);
+    DCHECK(e.good());
+
+    // Ignore root object, should it ever change.
+    if (e.IsRoot())
+      continue;
+    // Ignore non-bookmark objects.
+    if (!e.Get(syncable::IS_BOOKMARK_OBJECT))
+      continue;
+
+    if (exists_now && !existed_before)
+      change_buffer_.PushAddedItem(id);
+    else if (!exists_now && existed_before)
+      change_buffer_.PushDeletedItem(id);
+    else if (exists_now && existed_before && BookmarkPropertiesDiffer(*i, e))
+      change_buffer_.PushUpdatedItem(id, BookmarkPositionsDiffer(*i, e));
+  }
+}
+
+SyncManager::Status::Summary
+SyncManager::SyncInternal::ComputeAggregatedStatusSummary() {
+  switch (allstatus()->status().icon) {
+    case AllStatus::OFFLINE:
+      return Status::OFFLINE;
+    case AllStatus::OFFLINE_UNSYNCED:
+      return Status::OFFLINE_UNSYNCED;
+    case AllStatus::SYNCING:
+      return Status::SYNCING;
+    case AllStatus::READY:
+      return Status::READY;
+    case AllStatus::CONFLICT:
+      return Status::CONFLICT;
+    case AllStatus::OFFLINE_UNUSABLE:
+      return Status::OFFLINE_UNUSABLE;
+    default:
+      return Status::INVALID;
+  }
+}
+
+SyncManager::Status SyncManager::SyncInternal::ComputeAggregatedStatus() {
+  Status return_status =
+      { ComputeAggregatedStatusSummary(),
+        allstatus()->status().authenticated,
+        allstatus()->status().server_up,
+        allstatus()->status().server_reachable,
+        allstatus()->status().server_broken,
+        allstatus()->status().notifications_enabled,
+        allstatus()->status().notifications_received,
+        allstatus()->status().notifications_sent,
+        allstatus()->status().unsynced_count,
+        allstatus()->status().conflicting_count,
+        allstatus()->status().syncing,
+        allstatus()->status().initial_sync_ended,
+        allstatus()->status().syncer_stuck,
+        allstatus()->status().updates_available,
+        allstatus()->status().updates_received,
+        allstatus()->status().disk_full,
+        allstatus()->status().max_consecutive_errors};
+  return return_status;
+}
+
+void SyncManager::SyncInternal::HandleSyncerEvent(const SyncerEvent& event) {
+  if (!initialized()) {
+    // We get here if A) We have successfully authenticated at least once (
+    // because we attach HandleSyncerEvent only once we receive notification of
+    // successful authentication [locally or otherwise]), but B) the initial
+    // sync had not completed at that time.
+    if (SyncerStatus(event.last_session).IsShareUsable())
+      MarkAndNotifyInitializationComplete();
+    return;
+  }
+
+  if (!observer_)
+    return;
+
+  // Only send an event if this is due to a cycle ending and this cycle
+  // concludes a canonical "sync" process; that is, based on what is known
+  // locally we are "all happy" and up-to-date.  There may be new changes on
+  // the server, but we'll get them on a subsequent sync.
+  //
+  // Notifications are sent at the end of every sync cycle, regardless of
+  // whether we should sync again.
+  if (event.what_happened == SyncerEvent::SYNC_CYCLE_ENDED) {
+    if (!event.last_session->ShouldSyncAgain()) {
+      observer_->OnSyncCycleCompleted();
+    }
+
+    // TODO(chron): Consider changing this back to track ShouldSyncAgain
+    // Only notify peers if a commit has occurred and change the bookmark model.
+    if (event.last_session && event.last_session->items_committed()) {
+      notification_pending_ = true;
+    }
+
+    // SyncCycles are started by the following events: creation of the syncer,
+    // (re)connection to buzz, local changes, peer notifications of updates.
+    // Peers will be notified of changes made while there is no buzz connection
+    // immediately after a connection has been re-established.
+    // the next sync cycle.
+    // TODO(brg): Move this to TalkMediatorImpl as a SyncerThread event hook.
+    if (notification_pending_ && talk_mediator()) {
+        LOG(INFO) << "Sending XMPP notification...";
+        bool success = talk_mediator()->SendNotification();
+        if (success) {
+          notification_pending_ = false;
+        }
+    } else {
+      LOG(INFO) << "Didn't send XMPP notification!"
+                   << " event.last_session: " << event.last_session
+                   << " event.last_session->items_committed(): "
+                   << event.last_session->items_committed()
+                   << " talk_mediator(): " << talk_mediator();
+    }
+  }
+}
+
+void SyncManager::SyncInternal::HandleAuthWatcherEvent(
+    const AuthWatcherEvent& event) {
+  // We don't care about an authentication attempt starting event, and we
+  // don't want to reset our state to AUTH_PROBLEM_NONE because the fact that
+  // an _attempt_ is starting doesn't change the fact that we have an auth
+  // problem.
+  if (event.what_happened == AuthWatcherEvent::AUTHENTICATION_ATTEMPT_START)
+    return;
+  // We clear our last auth problem cache on new auth watcher events, and only
+  // set it to indicate a problem state for certain AuthWatcherEvent types.
+  auth_problem_ = AUTH_PROBLEM_NONE;
+  switch (event.what_happened) {
+    case AuthWatcherEvent::AUTH_SUCCEEDED:
+      // We now know the supplied username and password were valid. If this
+      // wasn't the first sync, authenticated_name should already be assigned.
+      if (username_for_share().empty()) {
+        browser_sync::ToPathString s(event.user_email);
+        if (s.good())
+          share_.authenticated_name = s.get_string16();
+      }
+
+      DCHECK(LowerCaseEqualsASCII(browser_sync::ToUTF8(
+          username_for_share()).get_string(),
+          StringToLowerASCII(event.user_email).c_str()))
+          << "username_for_share= "
+          << browser_sync::ToUTF8(username_for_share())
+          << ", event.user_email= " << event.user_email;
+
+      if (observer_)
+        observer_->OnAuthProblem(AUTH_PROBLEM_NONE);
+
+      // Hook up the DirectoryChangeEvent listener, HandleChangeEvent.
+      {
+        syncable::ScopedDirLookup lookup(dir_manager(), username_for_share());
+        if (!lookup.good()) {
+          DCHECK(false) << "ScopedDirLookup creation failed; unable to hook "
+                        << "up directory change event listener!";
+          return;
+        }
+        dir_change_hookup_.reset(NewEventListenerHookup(
+            lookup->changes_channel(), this,
+            &SyncInternal::HandleChangeEvent));
+
+        if (lookup->initial_sync_ended())
+          MarkAndNotifyInitializationComplete();
+      }
+      {
+        // Start watching the syncer channel directly here.
+        DCHECK(syncer_thread() != NULL);
+        syncer_event_.reset(NewEventListenerHookup(syncer_thread()->channel(),
+            this, &SyncInternal::HandleSyncerEvent));
+      }
+      return;
+    // Authentication failures translate to Status::AuthProblem events.
+    case AuthWatcherEvent::GAIA_AUTH_FAILED:     // Invalid GAIA credentials.
+    case AuthWatcherEvent::SERVICE_AUTH_FAILED:  // Expired GAIA credentials.
+      auth_problem_ = AUTH_PROBLEM_INVALID_GAIA_CREDENTIALS;
+      break;
+    case AuthWatcherEvent::SERVICE_USER_NOT_SIGNED_UP:
+      auth_problem_ = AUTH_PROBLEM_USER_NOT_SIGNED_UP;
+      break;
+    case AuthWatcherEvent::SERVICE_CONNECTION_FAILED:
+      auth_problem_ = AUTH_PROBLEM_CONNECTION_FAILED;
+      break;
+    default:  // We don't care about the many other AuthWatcherEvent types.
+      return;
+  }
+
+  // Fire notification that the status changed due to an authentication error.
+  if (observer_)
+    observer_->OnAuthProblem(auth_problem_);
+}
+
+SyncManager::Status::Summary SyncManager::GetStatusSummary() const {
+  return data_->ComputeAggregatedStatusSummary();
+}
+
+SyncManager::Status SyncManager::GetDetailedStatus() const {
+  return data_->ComputeAggregatedStatus();
+}
+
+SyncManager::SyncInternal* SyncManager::GetImpl() const { return data_; }
+
+void SyncManager::SaveChanges() {
+  data_->SaveChanges();
+}
+
+void SyncManager::SyncInternal::SaveChanges() {
+  syncable::ScopedDirLookup lookup(dir_manager(), username_for_share());
+  if (!lookup.good()) {
+    DCHECK(false) << "ScopedDirLookup creation failed; Unable to SaveChanges";
+    return;
+  }
+  lookup->SaveChanges();
+}
+
+void SyncManager::SetupForTestMode(const sync_char16* test_username) {
+  DCHECK(data_) << "SetupForTestMode requires initialization";
+  data_->SetupForTestMode(test_username);
+}
+
+void SyncManager::SyncInternal::SetupForTestMode(
+    const sync_char16* test_username) {
+  String16ToPathString(test_username, &share_.authenticated_name);
+
+  if (!dir_manager()->Open(username_for_share()))
+    DCHECK(false) << "Could not open directory when running in test mode";
+
+  // Hook up the DirectoryChangeEvent listener, HandleChangeEvent.
+  {
+    syncable::ScopedDirLookup lookup(dir_manager(), username_for_share());
+    if (!lookup.good()) {
+      DCHECK(false) << "ScopedDirLookup creation failed; unable to hook "
+                    << "up directory change event listener!";
+      return;
+    }
+    dir_change_hookup_.reset(NewEventListenerHookup(
+        lookup->changes_channel(), this,
+        &SyncInternal::HandleChangeEvent));
+  }
+  MarkAndNotifyInitializationComplete();
+}
+
+//////////////////////////////////////////////////////////////////////////
+// BaseTransaction member definitions
+BaseTransaction::BaseTransaction(UserShare* share)
+    : lookup_(NULL) {
+  DCHECK(share && share->dir_manager.get());
+  lookup_ = new syncable::ScopedDirLookup(share->dir_manager.get(),
+                                        share->authenticated_name);
+  if (!(lookup_->good()))
+    DCHECK(false) << "ScopedDirLookup failed on valid DirManager.";
+}
+BaseTransaction::~BaseTransaction() {
+  delete lookup_;
+}
+
+UserShare* SyncManager::GetUserShare() const {
+  DCHECK(data_->initialized()) << "GetUserShare requires initialization!";
+  return data_->GetUserShare();
+}
+
+}  // namespace sync_api
Property changes on: chrome\browser\sync\engine\syncapi.cc
___________________________________________________________________
Added: svn:eol-style
   + LF