So I have a custom subclass of OrmLiteSqliteOpenHelper. I want to use the ObjectCache interface to make sure I have identity-mapping from DB rows to in-memory objects, so I override getDao(...) as:
@Override
public <D extends Dao<T, ?>, T> D getDao(Class<T> arg0) throws SQLException {
D dao = super.getDao(arg0);
if (dao.getObjectCache() == null && !UNCACHED_CLASSES.contains(arg0))
dao.setObjectCache(InsightOpenHelperManager.sharedCache());
return dao;
}
My understanding is that super.getDao(Class<T> clazz) is basically doing a call to DaoManager.createDao(this.getConnectionSource(),clazz) behind the scenes, which should find a cached DAO if one exists. However…
final DatabaseHelper helpy = CustomOpenHelperManager.getHelper(StoreDatabaseHelper.class);
final CoreDao<Store, Integer> storeDao = helpy.getDao(Store.class);
DaoManager.registerDao(helpy.getConnectionSource(), storeDao);
final Dao<Store,Integer> testDao = DaoManager.createDao(helpy.getConnectionSource(), Store.class);
I would expect that (even w/o the registerDao(...) call) storeDao and testDao should be references to the same object. I see this in the Eclipse debugger, however:
Also, testDao‘s object cache is null.
Am I doing something wrong here? Is this a bug?
I do have a custom helper manager, but only because I needed to manage several databases. It’s just a hashmap of Class<? extends DatabaseHelper> keys to instances.
The reason I need my DAO cached is that I have several foreign collections that are eager and are being loaded by internally-generated DAOs that are not using my global cache and thus are being re-created independently for each collection.
As I was writing this up, I thought I could just have my overridden helpy.getDao(...) call through to DaoManager.createDao(...), but that results in the same thing: I still get a different DAO on the second call to createDao(...). This seems to me to be totally against the docs for DaoManager.
First, I thought it looked like registerDao(...) may be the culprit:
public static synchronized void registerDao(ConnectionSource connectionSource, Dao<?, ?> dao) {
if (connectionSource == null) {
throw new IllegalArgumentException("connectionSource argument cannot be null");
}
if (dao instanceof BaseDaoImpl) {
DatabaseTableConfig<?> tableConfig = ((BaseDaoImpl<?, ?>) dao).getTableConfig();
if (tableConfig != null) {
tableMap.put(new TableConfigConnectionSource(connectionSource, tableConfig), dao);
return;
}
}
classMap.put(new ClassConnectionSource(connectionSource, dao.getDataClass()), dao);
}
That return on line 230 of the source for DaoManager prevents the classMap from being updated (since I’m using the pregenerated config files?). When my code hits the second create call, it looks at the classMap first, and somehow (against my better understanding) finds a different copy of the DAO living there. Which is super weird, because stepping through the first create, I watched the classMap be initialized.
But where would a second DAO possibly come from?
Looking forward to Gray’s insight! 🙂
As @Ben mentioned, there is some internal DAO creation which is screwing things up but I think he may have uncovered a bug.
Under Android, ORMLite tries to use some magic reflection to build the DAOs given the horrible reflection performance under all but the most recent Android OS versions. Whenever the user asks for the DAO for class
Store(for example), the magic reflection fu is creating one DAO but internally it is using another one. I’ve created the follow bug:I changed the way the DAOs get created to do a better job of using the reflection output. The changes were pushed out in the 4.34. This release revamps (and simplifies) the internal DAO creation and caching. It should fix the issue.