def connect(self): """Connect to the database. Connects to the database only if the store has not already and provided that the store has a valid model. The default implementation of connect() is usually sufficient provided that subclasses have implemented newConnection(). """ assert self._model, 'Cannot connect:' \ ' No model has been attached to this store yet.' if not self._connected: self._connection = self.newConnection() self._connected = True self.readKlassIds() poolSize = self.setting('SQLConnectionPoolSize', 0) if poolSize: args = self._dbArgs.copy() self.augmentDatabaseArgs(args, pool=True) try: self._pool = DBPool(self.dbapiModule(), poolSize, **args) except TypeError: if 'database' in args: del args['database'] self._pool = DBPool(self.dbapiModule(), poolSize, **args) else: raise
def testDbPool(self): pool = DBPool(sqlite3, 10, database=':memory:') for _count in range(15): con = pool.connection() cursor = con.cursor() cursor.execute("select 1 union select 2 union select 3 order by 1") rows = cursor.fetchall() self.assertEqual(rows, [(1, ), (2, ), (3, )]) con.close()
def Test(iterations=15): try: dbapi_name = 'pgdb' dbapi = __import__(dbapi_name) pool = DBPool(dbapi, 10, database='template1') for i in range(iterations): db = pool.connection() cursor = db.cursor() cursor.execute("select datname from pg_database order by 1") r = cursor.fetchmany(5) r = ', '.join(map(lambda s: s[0], r)) print i, r db.close() except Exception: import traceback traceback.print_exc() print 'You need the pgdb adapter and a test database for this example.'
def connect(self): """ Connects to the database only if the store has not already and provided that the store has a valid model. The default implementation of connect() is usually sufficient provided that subclasses have implemented newConnection(). """ assert self._model, 'Cannot connect: No model has been attached to this store yet.' if not self._connected: self._connection = self.newConnection() self._connected = 1 self.readKlassIds() poolSize = self.setting('SQLConnectionPoolSize', 0) if poolSize: args = self._dbArgs.copy() if not args.get('db'): args['db'] = self._model.sqlDatabaseName() self._pool = DBPool(self.dbapiModule(), poolSize, **args)
class SQLObjectStore(ObjectStore): """ TO DO: * _sqlEcho should be accessible via a config file setting as stdout, stderr or a filename. For details on DB API 2.0, including the thread safety levels see: http://www.python.org/topics/database/DatabaseAPI-2.0.html """ ## Init ## def __init__(self, **kwargs): # @@ 2001-02-12 ce: We probably need a dictionary before kwargs for subclasses to pass to us in case they override __init__() and end up collecting kwargs themselves ObjectStore.__init__(self) self._dbArgs = kwargs self._connected = 0 self._sqlEcho = None self._sqlCount = 0 self._pool = None # an optional DBPool def modelWasSet(self): """ Performs additional set up of the store after the model is set, normally via setModel() or readModelFileNamed(). This includes checking that threading conditions are valid, and connecting to the database. """ ObjectStore.modelWasSet(self) # Check thread safety self._threadSafety = self.threadSafety() if self._threaded and self._threadSafety == 0: raise SQLObjectStoreThreadingError, 'Threaded is 1, but the DB API threadsafety is 0.' # Cache some settings self._markDeletes = self.setting('DeleteBehavior', 'delete') == 'mark' # Set up SQL echo self.setUpSQLEcho() # Set up attrs for caching for klass in self.model().allKlassesInOrder(): klass._getMethods = {} klass._setMethods = {} for attr in klass.allDataAttrs(): attr._sqlColumnName = None # Connect self.connect() def setUpSQLEcho(self): """ Sets up the SQL echoing/logging for the store according to the setting 'SQLLog'. See the User's Guide for more info. Invoked by modelWasSet(). """ setting = self.setting('SQLLog', None) if setting == None or setting == {}: self._sqlEcho = None else: filename = setting['File'] if filename == None: self._sqlEcho = None elif filename == 'stdout': self._sqlEcho = sys.stdout elif filename == 'stderr': self._sqlEcho = sys.stderr else: mode = setting.get('Mode', 'write') assert mode in ['write', 'append'] mode = mode[0] self._sqlEcho = open(filename, mode) ## Connecting to the db ## def isConnected(self): return self._connected def connect(self): """ Connects to the database only if the store has not already and provided that the store has a valid model. The default implementation of connect() is usually sufficient provided that subclasses have implemented newConnection(). """ assert self._model, 'Cannot connect: No model has been attached to this store yet.' if not self._connected: self._connection = self.newConnection() self._connected = 1 self.readKlassIds() poolSize = self.setting('SQLConnectionPoolSize', 0) if poolSize: args = self._dbArgs.copy() if not args.get('db'): args['db'] = self._model.sqlDatabaseName() self._pool = DBPool(self.dbapiModule(), poolSize, **args) def newConnection(self): """ Returns a DB API 2.0 connection. This is a utility method invoked by connect(). Subclasses should implement this, making use of self._dbArgs (a dictionary specifying host, username, etc.) as well as self._model.sqlDatabaseName(). Subclass responsibility. """ raise AbstractError, self.__class__ def readKlassIds(self): """ Reads the klass ids from the SQL database. Invoked by connect(). """ conn, cur = self.executeSQL('select id, name from _MKClassIds;') klassesById = {} for (id, name) in cur.fetchall(): assert id, "Id must be a non-zero int. id=%r, name=%r" % (id, name) klass = self._model.klass(name) klassesById[id] = klass klass.setId(id) self._klassesById = klassesById ## Changes ## def commitInserts(self, allThreads=0): unknownSerialNums = [] for object in self._newObjects.items(allThreads): self._insertObject(object, unknownSerialNums) for unknownInfo in unknownSerialNums: stmt = unknownInfo.updateStmt() self.executeSQL(stmt) self._newObjects.clear(allThreads) def _insertObject(self, object, unknownSerialNums): # New objects not in the persistent store have serial numbers less than 1 if object.serialNum() > 0: try: rep = repr(object) except: rep = '(repr exception)' assert object.serialNum() < 1, 'object=%s' % rep # SQL insert sql = object.sqlInsertStmt(unknownSerialNums) conn, cur = self.executeSQL(sql) # Get new id/serial num idNum = self.retrieveLastInsertId(conn, cur) # Update object object.setSerialNum(idNum) object.setKey(ObjectKey().initFromObject(object)) object.setChanged(0) # Update our object pool self._objects[object.key()] = object def retrieveLastInsertId(self, conn, cur): """ Returns the id (typically a 32-bit int) of the last INSERT operation by this connection. Used by commitInserts() to get the correct serial number for the last inserted object. Subclass responsibility. """ raise AbstractError, self.__class__ def commitUpdates(self, allThreads=0): for object in self._changedObjects.values(allThreads): sql = object.sqlUpdateStmt() self.executeSQL(sql) object.setChanged(0) self._changedObjects.clear(allThreads) def commitDeletions(self, allThreads=0): for object in self._deletedObjects.items(allThreads): sql = object.sqlDeleteStmt() self.executeSQL(sql) self._deletedObjects.clear(allThreads) ## Fetching ## def fetchObject(self, aClass, serialNum, default=NoDefault): """ Fetches a single object of a specific class and serial number. TheClass can be a Klass object (from the MiddleKit object model), the name of the class (e.g., a string) or a Python class. Raises an exception if theClass parameter is invalid, or the object cannot be located. """ klass = self._klassForClass(aClass) objects = self.fetchObjectsOfClass(klass, serialNum=serialNum, isDeep=0) count = len(objects) if count == 0: if default is NoDefault: raise UnknownObjectError, 'aClass = %r, serialNum = %r' % ( aClass, serialNum) else: return default else: assert count == 1 return objects[0] def fetchObjectsOfClass(self, aClass, clauses='', isDeep=1, refreshAttrs=1, serialNum=None): """ Fetches a list of objects of a specific class. The list may be empty if no objects are found. aClass can be a Klass object (from the MiddleKit object model), the name of the class (e.g., a string) or a Python class. The clauses argument can be any SQL clauses such as 'where x<5 order by x'. Obviously, these could be specific to your SQL database, thereby making your code non-portable. Use your best judgement. serialNum can be a specific serial number if you are looking for a specific object. If serialNum is provided, it overrides the clauses. You should label all arguments other than aClass: objs = store.fetchObjectsOfClass('Foo', clauses='where x<5') The reason for labeling is that this method is likely to undergo improvements in the future which could include additional arguments. No guarantees are made about the order of the arguments except that aClass will always be the first. Raises an exception if aClass parameter is invalid. """ klass = self._klassForClass(aClass) # Fetch objects of subclasses first, because the code below will be modifying clauses and serialNum deepObjs = [] if isDeep: for subklass in klass.subklasses(): deepObjs.extend( self.fetchObjectsOfClass(subklass, clauses, isDeep, refreshAttrs, serialNum)) # Now get objects of this exact class objs = [] if not klass.isAbstract(): fetchSQLStart = klass.fetchSQLStart() className = klass.name() if serialNum is not None: clauses = 'where %s=%d' % (klass.sqlIdName(), serialNum) if self._markDeletes: clauses = self.addDeletedToClauses(clauses) conn, cur = self.executeSQL(fetchSQLStart + clauses + ';') for row in cur.fetchall(): serialNum = row[0] key = ObjectKey().initFromClassNameAndSerialNum( className, serialNum) obj = self._objects.get(key, None) if obj is None: pyClass = klass.pyClass() obj = pyClass() assert isinstance( obj, MiddleObject ), 'Not a MiddleObject. obj = %r, type = %r, MiddleObject = %r' % ( obj, type(obj), MiddleObject) obj.readStoreData(self, row) obj.setKey(key) self._objects[key] = obj else: # Existing object if refreshAttrs: obj.readStoreData(self, row) objs.append(obj) objs.extend(deepObjs) return objs ## Klasses ## def klassForId(self, id): return self._klassesById[id] ## Self utility for SQL, connections, cursors, etc. ## def executeSQL(self, sql, connection=None): """ Executes the given SQL, connecting to the database for the first time if necessary. This method will also log the SQL to self._sqlEcho, if it is not None. Returns the connection and cursor used and relies on connectionAndCursor() to obtain these. Note that you can pass in a connection to force a particular one to be used. """ if aggressiveGC: import gc assert gc.isenabled() gc.collect() self._sqlCount += 1 if self._sqlEcho: timestamp = funcs.timestamp()['pretty'] self._sqlEcho.write('SQL %04i. %s %s\n' % (self._sqlCount, timestamp, sql)) self._sqlEcho.flush() # if self._sqlCount in (2, 3, 4): # import traceback as tb # tb.print_stack() conn, cur = self.connectionAndCursor(connection) self._executeSQL(cur, sql.strip()) return conn, cur def _executeSQL(self, cur, sql): """ Invokes execute on the cursor with the given SQL. This is a hook for subclasses that wish to influence this event. Invoked by executeSQL(). """ cur.execute(sql) def setSQLEcho(self, file): """ Sets a file to echo sql statements to, as sent through executeSQL(). None can be passed to turn echo off. """ self._sqlEcho = file def connectionAndCursor(self, connection=None): """ Returns the connection and cursor needed for executing SQL, taking into account factors such as setting('Threaded') and the threadsafety level of the DB API module. You can pass in a connection to force a particular one to be used. Uses newConnection() and connect(). """ if aggressiveGC: import gc assert gc.isenabled() gc.collect() if connection: conn = connection elif self._threaded: if self._pool: conn = self._pool.getConnection() elif self._threadSafety is 1: conn = self.newConnection() else: # safety = 2, 3 if not self._connected: self.connect() conn = self._connection else: # Non-threaded if not self._connected: self.connect() conn = self._connection cursor = conn.cursor() return conn, cursor def newConnection(self): """ Subclasses must override to return a newly created database connection. """ raise AbstractError, self.__class__ def threadSafety(self): return self.dbapiModule().threadsafety def addDeletedToClauses(self, clauses): """ Modify the given set of clauses so that it filters out records with non-NULL deleted field """ clauses = clauses.strip() if clauses.lower().startswith('where'): orderByIndex = clauses.lower().find('order by') if orderByIndex == -1: where = clauses[5:] orderBy = '' else: where = clauses[5:orderByIndex] orderBy = clauses[orderByIndex:] return 'where deleted is null and (%s) %s' % (where, orderBy) else: return 'where deleted is null %s' % clauses ## Obj refs ## def fetchObjRef(self, objRef): """ Given an unarchived object reference, this method returns the actual object for it (or None if the reference is NULL or dangling). While this method assumes that obj refs are stored as 64-bit numbers containing the class id and object serial number, subclasses are certainly able to override that assumption by overriding this method. """ assert type(objRef) is LongType, 'type=%r, objRef=%r' % (type(objRef), objRef) if objRef == 0: return None else: klassId, serialNum = objRefSplit(objRef) if klassId == 0 or serialNum == 0: # invalid! we don't use 0 serial numbers return self.objRefZeroSerialNum(objRef) klass = self.klassForId(klassId) # Check if we already have this in memory first key = ObjectKey() key.initFromClassNameAndSerialNum(klass.name(), serialNum) obj = self._objects.get(key, None) if obj: return obj clauses = 'where %s=%d' % (klass.sqlIdName(), serialNum) objs = self.fetchObjectsOfClass(klass, clauses, isDeep=0) if len(objs) == 1: return objs[0] elif len(objs) > 1: raise ValueError, 'Multiple objects.' # @@ 2000-11-22 ce: expand the msg with more information else: return self.objRefDangles(objRef) def objRefZeroSerialNum(self, objRef): """ Invoked by fetchObjRef() if either the class or object serial number is 0. """ raise ObjRefZeroSerialNumError, objRefSplit(objRef) def objRefDangles(self, objRef): """ Invoked by fetchObjRef() if there is no possible target object for the given objRef, e.g., a dangling reference. This method invokes self.warning() and includes the objRef as decimal, hexadecimal and class:obj numbers. """ raise ObjRefDanglesError, objRefSplit(objRef) ## Debugging ## def dumpTables(self, out=None): if out is None: out = sys.stdout out.write('DUMPING TABLES\n') out.write('BEGIN\n') for klass in self.model().klasses().values(): out.write(klass.name() + '\n') self.executeSQL('select * from %s;' % klass.name()) out.write(str(self._cursor.fetchall())) out.write('\n') out.write('END\n') def dumpKlassIds(self, out=None): if out is None: out = sys.stdout wr = out.write('DUMPING KLASS IDs\n') for klass in self.model().klasses().values(): out.write('%25s %2i\n' % (klass.name(), klass.id())) out.write('\n')
class SQLObjectStore(ObjectStore): """The MiddleKit SQL Object Store. TO DO: * _sqlEcho should be accessible via a config file setting as stdout, stderr or a filename. For details on DB API 2.0, including the thread safety levels see: http://www.python.org/topics/database/DatabaseAPI-2.0.html """ ## Init ## def __init__(self, **kwargs): # @@ 2001-02-12 ce: We probably need a dictionary before kwargs # for subclasses to pass to us in case they override __init__() # and end up collecting kwargs themselves ObjectStore.__init__(self) self._dbArgs = kwargs self._connected = False self._commited = False self._sqlEcho = None self._sqlCount = 0 self._pool = None # an optional DBPool def modelWasSet(self): """Perform additional set up of the store after the model is set. Performs additional set up of the store after the model is set, normally via setModel() or readModelFileNamed(). This includes checking that threading conditions are valid, and connecting to the database. """ ObjectStore.modelWasSet(self) # Check thread safety self._threadSafety = self.threadSafety() if self._threaded and self._threadSafety == 0: raise SQLObjectStoreThreadingError('Threaded is True,' ' but the DB API threadsafety is 0.') # Cache some settings self._markDeletes = self.setting('DeleteBehavior', 'delete') == 'mark' # Set up SQL echo self.setUpSQLEcho() # Set up attrs for caching for klass in self.model().allKlassesInOrder(): klass._getMethods = {} klass._setMethods = {} for attr in klass.allDataAttrs(): attr._sqlColumnName = None attr._sqlColumnNames = None # use dbargs from settings file as defaults # (args passed to __init__ take precedence) args = self._dbArgs self._dbArgs = self.setting('DatabaseArgs', {}) self._dbArgs.update(args) # print 'dbArgs = %s' % self._dbArgs # Connect self.connect() def setUpSQLEcho(self): """Set up the SQL echoing/logging for the store. The logging is set up according to the setting 'SQLLog'. See the User's Guide for more info. Invoked by modelWasSet(). """ setting = self.setting('SQLLog', None) if setting is None or setting == {}: self._sqlEcho = None else: filename = setting['File'] if filename is None: self._sqlEcho = None elif filename == 'stdout': self._sqlEcho = sys.stdout elif filename == 'stderr': self._sqlEcho = sys.stderr else: mode = setting.get('Mode', 'write') assert mode in ['write', 'append'] mode = mode[0] self._sqlEcho = open(filename, mode) ## Connecting to the db ## def isConnected(self): return self._connected def connect(self): """Connect to the database. Connects to the database only if the store has not already and provided that the store has a valid model. The default implementation of connect() is usually sufficient provided that subclasses have implemented newConnection(). """ assert self._model, 'Cannot connect:' \ ' No model has been attached to this store yet.' if not self._connected: self._connection = self.newConnection() self._connected = True self.readKlassIds() poolSize = self.setting('SQLConnectionPoolSize', 0) if poolSize: args = self._dbArgs.copy() self.augmentDatabaseArgs(args, pool=True) try: self._pool = DBPool(self.dbapiModule(), poolSize, **args) except TypeError: if 'database' in args: del args['database'] self._pool = DBPool(self.dbapiModule(), poolSize, **args) else: raise def augmentDatabaseArgs(self, args, pool=False): # give subclasses the opportunity to add or change # database arguments pass def newConnection(self): """Return a DB API 2.0 connection. This is a utility method invoked by connect(). Subclasses should implement this, making use of self._dbArgs (a dictionary specifying host, username, etc.) as well as self._model.sqlDatabaseName(). Subclass responsibility. """ raise AbstractError(self.__class__) def readKlassIds(self): """Read the klass ids from the SQL database. Invoked by connect().""" conn, cur = self.executeSQL('select id, name from _MKClassIds;') try: klassesById = {} for klassId, name in cur.fetchall(): assert klassId, 'Id must be a non-zero int.' \ ' id=%r, name=%r' % (klassId, name) try: klass = self._model.klass(name) except KeyError: filename = self._model.filename() msg = ('%s The database has a class id for %r in the' ' _MKClassIds table, but no such class exists in' ' the model %s. The model and the db are not in sync.' % (name, name, filename)) raise KeyError(msg) klassesById[klassId] = klass klass.setId(klassId) finally: self.doneWithConnection(conn) self._klassesById = klassesById ## Changes ## def commitInserts(self, allThreads=False): unknownSerialNums = [] # @@ ... sort here for dependency order for obj in self._newObjects.items(allThreads): self._insertObject(obj, unknownSerialNums) conn = None try: for unknownInfo in unknownSerialNums: stmt = unknownInfo.updateStmt() conn, cur = self.executeSQL(stmt, conn) finally: self.doneWithConnection(conn) self._newObjects.clear(allThreads) def _insertObject(self, obj, unknownSerialNums): # New objects not in the persistent store have serial numbers less than 1 if obj.serialNum() > 0: try: rep = repr(obj) except Exception: rep = '(repr exception)' assert obj.serialNum() < 1, 'obj=%s' % rep # try to get the next ID (if database supports this) idNum = self.retrieveNextInsertId(obj.klass()) # SQL insert sql = obj.sqlInsertStmt(unknownSerialNums, idNum) conn, cur = self.executeSQL(sql) try: # Get new id/serial num if idNum is None: idNum = self.retrieveLastInsertId(conn, cur) # Update object obj.setSerialNum(idNum) obj.setKey(ObjectKey().initFromObject(obj)) obj.setChanged(False) # Update our object pool self._objects[obj.key()] = obj finally: self.doneWithConnection(conn) def retrieveNextInsertId(self, klass): """Return the id for the next new object of this class. Databases which cannot determine the id until the object has been added return None, signifying that retrieveLastInsertId should be called to get the id after the insert has been made. """ return None def retrieveLastInsertId(self, conn, cur): """Return the id of the last INSERT operation by this connection. This id is typically a 32-bit int. Used by commitInserts() to get the correct serial number for the last inserted object. """ return cur.lastrowid def commitUpdates(self, allThreads=False): conn = None try: for obj in self._changedObjects.values(allThreads): sql = obj.sqlUpdateStmt() conn, cur = self.executeSQL(sql, conn) obj.setChanged(False) finally: self.doneWithConnection(conn) self._changedObjects.clear(allThreads) def commitDeletions(self, allThreads=False): conn = None try: for obj in self._deletedObjects.items(allThreads): sql = obj.sqlDeleteStmt() conn, cur = self.executeSQL(sql, conn) conn.commit() finally: self.doneWithConnection(conn) self._deletedObjects.clear(allThreads) ## Fetching ## def fetchObject(self, aClass, serialNum, default=NoDefault): """Fetch a single object of a specific class and serial number. aClass can be a Klass object (from the MiddleKit object model), the name of the class (e.g., a string) or a Python class. Raises an exception if aClass parameter is invalid, or the object cannot be located. """ klass = self._klassForClass(aClass) objects = self.fetchObjectsOfClass(klass, serialNum=serialNum, isDeep=False) count = len(objects) if count == 0: if default is NoDefault: raise UnknownObjectError('aClass = %r, serialNum = %r' % (aClass, serialNum)) else: return default else: assert count == 1 return objects[0] def fetchObjectsOfClass(self, aClass, clauses='', isDeep=True, refreshAttrs=True, serialNum=None): """Fetch a list of objects of a specific class. The list may be empty if no objects are found. aClass can be a Klass object (from the MiddleKit object model), the name of the class (e.g., a string) or a Python class. The clauses argument can be any SQL clauses such as 'where x<5 order by x'. Obviously, these could be specific to your SQL database, thereby making your code non-portable. Use your best judgement. serialNum can be a specific serial number if you are looking for a specific object. If serialNum is provided, it overrides the clauses. You should label all arguments other than aClass: objs = store.fetchObjectsOfClass('Foo', clauses='where x<5') The reason for labeling is that this method is likely to undergo improvements in the future which could include additional arguments. No guarantees are made about the order of the arguments except that aClass will always be the first. Raises an exception if aClass parameter is invalid. """ klass = self._klassForClass(aClass) # Fetch objects of subclasses first, because the code below # will be modifying clauses and serialNum deepObjs = [] if isDeep: for subklass in klass.subklasses(): deepObjs.extend(self.fetchObjectsOfClass( subklass, clauses, isDeep, refreshAttrs, serialNum)) # Now get objects of this exact class objs = [] if not klass.isAbstract(): fetchSQLStart = klass.fetchSQLStart() className = klass.name() if serialNum is not None: serialNum = int(serialNum) # make sure it's a valid int clauses = 'where %s=%d' % (klass.sqlSerialColumnName(), serialNum) if self._markDeletes: clauses = self.addDeletedToClauses(clauses) conn, cur = self.executeSQL(fetchSQLStart + clauses + ';') try: for row in cur.fetchall(): serialNum = row[0] key = ObjectKey().initFromClassNameAndSerialNum(className, serialNum) obj = self._objects.get(key) if obj is None: pyClass = klass.pyClass() obj = pyClass() assert isinstance(obj, MiddleObject), ( 'Not a MiddleObject. obj = %r, type = %r, MiddleObject = %r' % (obj, type(obj), MiddleObject)) obj.readStoreData(self, row) obj.setKey(key) self._objects[key] = obj else: # Existing object if refreshAttrs: obj.readStoreData(self, row) objs.append(obj) finally: self.doneWithConnection(conn) objs.extend(deepObjs) return objs def refreshObject(self, obj): assert obj.store() is self return self.fetchObject(obj.klass(), obj.serialNum()) ## Klasses ## def klassForId(self, id): return self._klassesById[id] ## Self utility for SQL, connections, cursors, etc. ## def executeSQL(self, sql, connection=None, commit=False): """Execute the given SQL. This will connect to the database for the first time if necessary. This method will also log the SQL to self._sqlEcho, if it is not None. Returns the connection and cursor used and relies on connectionAndCursor() to obtain these. Note that you can pass in a connection to force a particular one to be used and a flag to commit immediately. """ sql = str(sql) # Excel-based models yield Unicode strings which some db modules don't like sql = sql.strip() if aggressiveGC: import gc assert gc.isenabled() gc.collect() self._sqlCount += 1 if self._sqlEcho: timestamp = funcs.timestamp()['pretty'] self._sqlEcho.write('SQL %04i. %s %s\n' % (self._sqlCount, timestamp, sql)) self._sqlEcho.flush() conn, cur = self.connectionAndCursor(connection) self._executeSQL(cur, sql) if commit: conn.commit() return conn, cur def _executeSQL(self, cur, sql): """Invoke execute on the cursor with the given SQL. This is a hook for subclasses that wish to influence this event. Invoked by executeSQL(). """ cur.execute(sql) def executeSQLTransaction(self, transaction, connection=None, commit=True): """Execute the given sequence of SQL statements and commit as transaction.""" if isinstance(transaction, basestring): transaction = [transaction] try: for sql in transaction: if connection: self.executeSQL(sql, connection) else: connection, cur = self.executeSQL(sql) except Exception: if connection and commit: connection.rollback() raise if transaction and connection and commit: try: connection.commit() except Exception: connection.rollback() raise return connection def executeSQLScript(self, sql, connection=None): """Execute the given SQL script. This uses the nonstandard executescript() method as provided by the PySQLite adapter. """ sql = str(sql).strip() if not connection: connection = self.newConnection() if not hasattr(connection, 'executescript'): raise AttributeError( 'Script execution not supported by database adapter.') return connection.executescript(sql) def setSQLEcho(self, file): """Set a file to echo sql statements to, as sent through executeSQL(). None can be passed to turn echo off. """ self._sqlEcho = file def connectionAndCursor(self, connection=None): """Return the connection and cursor needed for executing SQL. Takes into account factors such as setting('Threaded') and the threadsafety level of the DB API module. You can pass in a connection to force a particular one to be used. Uses newConnection() and connect(). """ if aggressiveGC: import gc assert gc.isenabled() gc.collect() if connection: conn = connection elif self._threaded: if self._pool: conn = self._pool.connection() elif self._threadSafety == 1: conn = self.newConnection() else: # safety = 2, 3 if not self._connected: self.connect() conn = self._connection else: # Non-threaded if not self._connected: self.connect() conn = self._connection cursor = conn.cursor() return conn, cursor def threadSafety(self): """Return the threadsafety of the DB API module.""" return self.dbapiModule().threadsafety def dbapiVersion(self): """Return the version of the DB API module.""" module = self.dbapiModule() return '%s %s' % (module.__name__, module.version) def dbVersion(self): """Return the database version. Subclass responsibility. """ raise AbstractError(self.__class__) def addDeletedToClauses(self, clauses): """Modify the given set of clauses so that it filters out records with non-NULL deleted field.""" clauses = clauses.strip() if clauses.lower().startswith('where'): where = clauses[5:] orderByIndex = where.lower().find('order by') if orderByIndex < 0: orderBy = '' else: where, orderBy = where[:orderByIndex], where[orderByIndex:] return 'where deleted is null and (%s) %s' % (where, orderBy) else: return 'where deleted is null %s' % clauses ## Obj refs ## def fetchObjRef(self, objRef): """Fetch referenced object. Given an unarchived object reference, this method returns the actual object for it (or None if the reference is NULL or dangling). While this method assumes that obj refs are stored as 64-bit numbers containing the class id and object serial number, subclasses are certainly able to override that assumption by overriding this method. """ assert isinstance(objRef, long), 'type=%r, objRef=%r' % (type(objRef), objRef) if objRef == 0: return None else: klassId, serialNum = objRefSplit(objRef) if klassId == 0 or serialNum == 0: # invalid! we don't use 0 serial numbers return self.objRefZeroSerialNum(objRef) klass = self.klassForId(klassId) # Check if we already have this in memory first key = ObjectKey() key.initFromClassNameAndSerialNum(klass.name(), serialNum) obj = self._objects.get(key) if obj: return obj clauses = 'where %s=%d' % (klass.sqlSerialColumnName(), serialNum) objs = self.fetchObjectsOfClass(klass, clauses, isDeep=False) if len(objs) == 1: return objs[0] elif len(objs) > 1: # @@ 2000-11-22 ce: expand the msg with more information raise ValueError('Multiple objects.') else: return self.objRefDangles(objRef) def objRefInMem(self, objRef): """Return referenced object in memory. Returns the object corresponding to the given objref if and only if it has been loaded into memory. If the object has never been fetched from the database, None is returned. """ assert isinstance(objRef, long), 'type=%r, objRef=%r' % (type(objRef), objRef) if objRef == 0: return 0 else: klassId, serialNum = objRefSplit(objRef) if klassId == 0 or serialNum == 0: # invalid! we don't use 0 serial numbers return self.objRefZeroSerialNum(objRef) klass = self.klassForId(klassId) # return whether we have this object in memory key = ObjectKey() key.initFromClassNameAndSerialNum(klass.name(), serialNum) return self._objects.get(key) def objRefZeroSerialNum(self, objRef): """Raise serial number zero error. Invoked by fetchObjRef() if either the class or the object serial number is zero. """ raise ObjRefZeroSerialNumError(objRefSplit(objRef)) def objRefDangles(self, objRef): """Raise dangling reference error. Invoked by fetchObjRef() if there is no possible target object for the given objRef. E.g., this can happen for a dangling reference. This method invokes self.warning() and includes the objRef as decimal, hexadecimal and class:obj numbers. """ raise ObjRefDanglesError(objRefSplit(objRef)) ## Special Cases ## def filterDateTimeDelta(self, dtd): """Adapt DateTimeDeltas. Some databases have no TIME type and therefore will not return DateTimeDeltas. This utility method is overridden by subclasses as appropriate and invoked by the test suite. """ return dtd def sqlNowCall(self): """Get current DateTime.""" return 'CURRENT_TIMESTAMP' ## Self util ## def doneWithConnection(self, conn): """Invoked by self when a connection is no longer needed. The default behavior is to commit and close the connection. """ if conn is not None: # Starting with 1.2.0, MySQLdb disables autocommit by default, # as required by the DB-API standard (PEP-249). If you are using # InnoDB tables or some other type of transactional table type, # you'll need to do connection.commit() before closing the connection, # or else none of your changes will be written to the database. try: conn.commit() except Exception: pass conn.close() ## Debugging ## def dumpTables(self, out=None): if out is None: out = sys.stdout out.write('DUMPING TABLES\n') out.write('BEGIN\n') conn = None try: for klass in self.model().klasses().values(): out.write(klass.name() + '\n') conn, cur = self.executeSQL('select * from %s;' % klass.name(), conn) out.write(str(self._cursor.fetchall())) out.write('\n') finally: self.doneWithConnection(conn) out.write('END\n') def dumpKlassIds(self, out=None): if out is None: out = sys.stdout wr = out.write('DUMPING KLASS IDs\n') for klass in self.model().klasses().values(): out.write('%25s %2i\n' % (klass.name(), klass.id())) out.write('\n') def dumpObjectStore(self, out=None, progress=False): if out is None: out = sys.stdout for klass in self.model().klasses().values(): if progress: sys.stderr.write(".") out.write('%s objects\n' % (klass.name())) attrs = [attr for attr in klass.allAttrs() if attr.hasSQLColumn()] colNames = [attr.name() for attr in attrs] colNames.insert(0, klass.sqlSerialColumnName()) out.write(CSVJoiner.joinCSVFields(colNames) + "\n") # write out a line for each object in this class objlist = self.fetchObjectsOfClass(klass.name(), isDeep=False) for obj in objlist: fields = [] fields.append(str(obj.serialNum())) for attr in attrs: # jdh 2003-03-07: if the attribute is a dangling object reference, the value # will be None. This means that dangling references will _not_ be remembered # across dump/generate/create/insert procedures. method = getattr(obj, attr.pyGetName()) value = method() if value is None: fields.append('') elif isinstance(value, MiddleObject): fields.append('%s.%d' % (value.klass().name(), value.serialNum())) else: fields.append(str(value)) out.write(CSVJoiner.joinCSVFields(fields).replace('\r', '\\r')) out.write('\n') out.write('\n') out.write('\n') if progress: sys.stderr.write("\n")
class SQLObjectStore(ObjectStore): """The MiddleKit SQL Object Store. TO DO: * _sqlEcho should be accessible via a config file setting as stdout, stderr or a filename. For details on DB API 2.0, including the thread safety levels see: http://www.python.org/topics/database/DatabaseAPI-2.0.html """ ## Init ## def __init__(self, **kwargs): # @@ 2001-02-12 ce: We probably need a dictionary before kwargs # for subclasses to pass to us in case they override __init__() # and end up collecting kwargs themselves ObjectStore.__init__(self) self._dbArgs = kwargs self._connected = False self._commited = False self._sqlEcho = None self._sqlCount = 0 self._pool = None # an optional DBPool def modelWasSet(self): """Perform additional set up of the store after the model is set. Performs additional set up of the store after the model is set, normally via setModel() or readModelFileNamed(). This includes checking that threading conditions are valid, and connecting to the database. """ ObjectStore.modelWasSet(self) # Check thread safety self._threadSafety = self.threadSafety() if self._threaded and self._threadSafety == 0: raise SQLObjectStoreThreadingError('Threaded is True,' ' but the DB API threadsafety is 0.') # Cache some settings self._markDeletes = self.setting('DeleteBehavior', 'delete') == 'mark' # Set up SQL echo self.setUpSQLEcho() # Set up attrs for caching for klass in self.model().allKlassesInOrder(): klass._getMethods = {} klass._setMethods = {} for attr in klass.allDataAttrs(): attr._sqlColumnName = None attr._sqlColumnNames = None # use dbargs from settings file as defaults # (args passed to __init__ take precedence) args = self._dbArgs self._dbArgs = self.setting('DatabaseArgs', {}) self._dbArgs.update(args) # print 'dbArgs = %s' % self._dbArgs # Connect self.connect() def setUpSQLEcho(self): """Set up the SQL echoing/logging for the store. The logging is set up according to the setting 'SQLLog'. See the User's Guide for more info. Invoked by modelWasSet(). """ setting = self.setting('SQLLog', None) if setting is None or setting == {}: self._sqlEcho = None else: filename = setting['File'] if filename is None: self._sqlEcho = None elif filename == 'stdout': self._sqlEcho = sys.stdout elif filename == 'stderr': self._sqlEcho = sys.stderr else: mode = setting.get('Mode', 'write') assert mode in ['write', 'append'] mode = mode[0] self._sqlEcho = open(filename, mode) ## Connecting to the db ## def isConnected(self): return self._connected def connect(self): """Connect to the database. Connects to the database only if the store has not already and provided that the store has a valid model. The default implementation of connect() is usually sufficient provided that subclasses have implemented newConnection(). """ assert self._model, 'Cannot connect:' \ ' No model has been attached to this store yet.' if not self._connected: self._connection = self.newConnection() self._connected = True self.readKlassIds() poolSize = self.setting('SQLConnectionPoolSize', 0) if poolSize: args = self._dbArgs.copy() self.augmentDatabaseArgs(args, pool=True) try: self._pool = DBPool(self.dbapiModule(), poolSize, **args) except TypeError: if 'database' in args: del args['database'] self._pool = DBPool(self.dbapiModule(), poolSize, **args) else: raise def augmentDatabaseArgs(self, args, pool=False): # give subclasses the opportunity to add or change # database arguments pass def newConnection(self): """Return a DB API 2.0 connection. This is a utility method invoked by connect(). Subclasses should implement this, making use of self._dbArgs (a dictionary specifying host, username, etc.) as well as self._model.sqlDatabaseName(). Subclass responsibility. """ raise AbstractError(self.__class__) def readKlassIds(self): """Read the klass ids from the SQL database. Invoked by connect().""" conn, cur = self.executeSQL('select id, name from _MKClassIds;') try: klassesById = {} for klassId, name in cur.fetchall(): assert klassId, 'Id must be a non-zero int.' \ ' id=%r, name=%r' % (klassId, name) try: klass = self._model.klass(name) except KeyError: filename = self._model.filename() msg = ('%s The database has a class id for %r in the' ' _MKClassIds table, but no such class exists in' ' the model %s. The model and the db are not in sync.' % (name, name, filename)) raise KeyError(msg) klassesById[klassId] = klass klass.setId(klassId) finally: self.doneWithConnection(conn) self._klassesById = klassesById ## Changes ## def commitInserts(self, allThreads=False): unknownSerialNums = [] # @@ ... sort here for dependency order for obj in self._newObjects.items(allThreads): self._insertObject(obj, unknownSerialNums) conn = None try: for unknownInfo in unknownSerialNums: stmt = unknownInfo.updateStmt() conn, cur = self.executeSQL(stmt, conn) finally: self.doneWithConnection(conn) self._newObjects.clear(allThreads) def _insertObject(self, obj, unknownSerialNums): # New objects not in the persistent store have serial numbers less than 1 if obj.serialNum() > 0: try: rep = repr(obj) except Exception: rep = '(repr exception)' assert obj.serialNum() < 1, 'obj=%s' % rep # try to get the next ID (if database supports this) idNum = self.retrieveNextInsertId(obj.klass()) # SQL insert sql = obj.sqlInsertStmt(unknownSerialNums, idNum) conn, cur = self.executeSQL(sql) try: # Get new id/serial num if idNum is None: idNum = self.retrieveLastInsertId(conn, cur) # Update object obj.setSerialNum(idNum) obj.setKey(ObjectKey().initFromObject(obj)) obj.setChanged(False) # Update our object pool self._objects[obj.key()] = obj finally: self.doneWithConnection(conn) def retrieveNextInsertId(self, klass): """Return the id for the next new object of this class. Databases which cannot determine the id until the object has been added return None, signifying that retrieveLastInsertId should be called to get the id after the insert has been made. """ return None def retrieveLastInsertId(self, conn, cur): """Return the id of the last INSERT operation by this connection. This id is typically a 32-bit int. Used by commitInserts() to get the correct serial number for the last inserted object. """ return cur.lastrowid def commitUpdates(self, allThreads=False): conn = None try: for obj in self._changedObjects.values(allThreads): sql = obj.sqlUpdateStmt() conn, cur = self.executeSQL(sql, conn) obj.setChanged(False) finally: self.doneWithConnection(conn) self._changedObjects.clear(allThreads) def commitDeletions(self, allThreads=False): conn = None try: for obj in self._deletedObjects.items(allThreads): sql = obj.sqlDeleteStmt() conn, cur = self.executeSQL(sql, conn) conn.commit() finally: self.doneWithConnection(conn) self._deletedObjects.clear(allThreads) ## Fetching ## def fetchObject(self, aClass, serialNum, default=NoDefault): """Fetch a single object of a specific class and serial number. aClass can be a Klass object (from the MiddleKit object model), the name of the class (e.g., a string) or a Python class. Raises an exception if aClass parameter is invalid, or the object cannot be located. """ klass = self._klassForClass(aClass) objects = self.fetchObjectsOfClass(klass, serialNum=serialNum, isDeep=False) count = len(objects) if count == 0: if default is NoDefault: raise UnknownObjectError('aClass = %r, serialNum = %r' % (aClass, serialNum)) else: return default else: assert count == 1 return objects[0] def fetchObjectsOfClass(self, aClass, clauses='', isDeep=True, refreshAttrs=True, serialNum=None, clausesArgs=None): """Fetch a list of objects of a specific class. The list may be empty if no objects are found. aClass can be a Klass object (from the MiddleKit object model), the name of the class (e.g., a string) or a Python class. The clauses argument can be any SQL clauses such as 'where x<5 order by x'. Obviously, these could be specific to your SQL database, thereby making your code non-portable. Use your best judgement. serialNum can be a specific serial number if you are looking for a specific object. If serialNum is provided, it overrides the clauses. You should label all arguments other than aClass: objs = store.fetchObjectsOfClass('Foo', clauses='where x<5') The reason for labeling is that this method is likely to undergo improvements in the future which could include additional arguments. No guarantees are made about the order of the arguments except that aClass will always be the first. Raises an exception if aClass parameter is invalid. """ klass = self._klassForClass(aClass) # Fetch objects of subclasses first, because the code below # will be modifying clauses and serialNum deepObjs = [] if isDeep: for subklass in klass.subklasses(): deepObjs.extend(self.fetchObjectsOfClass( subklass, clauses, isDeep, refreshAttrs, serialNum, clausesArgs)) # Now get objects of this exact class objs = [] if not klass.isAbstract(): fetchSQLStart = klass.fetchSQLStart() className = klass.name() if serialNum is not None: serialNum = int(serialNum) # make sure it's a valid int clauses = 'where %s=%d' % (klass.sqlSerialColumnName(), serialNum) if self._markDeletes: clauses = self.addDeletedToClauses(clauses) conn, cur = self.executeSQL(fetchSQLStart + clauses + ';', clausesArgs=clausesArgs) try: for row in cur.fetchall(): serialNum = row[0] key = ObjectKey().initFromClassNameAndSerialNum(className, serialNum) obj = self._objects.get(key) if obj is None: pyClass = klass.pyClass() obj = pyClass() assert isinstance(obj, MiddleObject), ( 'Not a MiddleObject. obj = %r, type = %r, MiddleObject = %r' % (obj, type(obj), MiddleObject)) obj.readStoreData(self, row) obj.setKey(key) self._objects[key] = obj else: # Existing object if refreshAttrs: obj.readStoreData(self, row) objs.append(obj) finally: self.doneWithConnection(conn) objs.extend(deepObjs) return objs def refreshObject(self, obj): assert obj.store() is self return self.fetchObject(obj.klass(), obj.serialNum()) ## Klasses ## def klassForId(self, id): return self._klassesById[id] ## Self utility for SQL, connections, cursors, etc. ## def executeSQL(self, sql, connection=None, commit=False, clausesArgs=None): """Execute the given SQL. This will connect to the database for the first time if necessary. This method will also log the SQL to self._sqlEcho, if it is not None. Returns the connection and cursor used and relies on connectionAndCursor() to obtain these. Note that you can pass in a connection to force a particular one to be used and a flag to commit immediately. """ sql = str(sql) # Excel-based models yield Unicode strings which some db modules don't like sql = sql.strip() if aggressiveGC: import gc assert gc.isenabled() gc.collect() self._sqlCount += 1 if self._sqlEcho: timestamp = funcs.timestamp()['pretty'] self._sqlEcho.write('SQL %04i. %s %s\n' % (self._sqlCount, timestamp, sql)) self._sqlEcho.flush() conn, cur = self.connectionAndCursor(connection) self._executeSQL(cur, sql, clausesArgs) if commit: conn.commit() return conn, cur def _executeSQL(self, cur, sql, clausesArgs=None): """Invoke execute on the cursor with the given SQL. This is a hook for subclasses that wish to influence this event. Invoked by executeSQL(). """ cur.execute(sql, clausesArgs) def executeSQLTransaction(self, transaction, connection=None, commit=True): """Execute the given sequence of SQL statements and commit as transaction.""" if isinstance(transaction, basestring): transaction = [transaction] try: for sql in transaction: if connection: self.executeSQL(sql, connection) else: connection, cur = self.executeSQL(sql) except Exception: if connection and commit: connection.rollback() raise if transaction and connection and commit: try: connection.commit() except Exception: connection.rollback() raise return connection def executeSQLScript(self, sql, connection=None): """Execute the given SQL script. This uses the nonstandard executescript() method as provided by the PySQLite adapter. """ sql = str(sql).strip() if not connection: connection = self.newConnection() if not hasattr(connection, 'executescript'): raise AttributeError( 'Script execution not supported by database adapter.') return connection.executescript(sql) def setSQLEcho(self, file): """Set a file to echo sql statements to, as sent through executeSQL(). None can be passed to turn echo off. """ self._sqlEcho = file def connectionAndCursor(self, connection=None): """Return the connection and cursor needed for executing SQL. Takes into account factors such as setting('Threaded') and the threadsafety level of the DB API module. You can pass in a connection to force a particular one to be used. Uses newConnection() and connect(). """ if aggressiveGC: import gc assert gc.isenabled() gc.collect() if connection: conn = connection elif self._threaded: if self._pool: conn = self._pool.connection() elif self._threadSafety == 1: conn = self.newConnection() else: # safety = 2, 3 if not self._connected: self.connect() conn = self._connection else: # Non-threaded if not self._connected: self.connect() conn = self._connection cursor = conn.cursor() return conn, cursor def threadSafety(self): """Return the threadsafety of the DB API module.""" return self.dbapiModule().threadsafety def dbapiVersion(self): """Return the version of the DB API module.""" module = self.dbapiModule() return '%s %s' % (module.__name__, module.version) def dbVersion(self): """Return the database version. Subclass responsibility. """ raise AbstractError(self.__class__) def addDeletedToClauses(self, clauses): """Modify the given set of clauses so that it filters out records with non-NULL deleted field.""" clauses = clauses.strip() if clauses.lower().startswith('where'): where = clauses[5:] orderByIndex = where.lower().find('order by') if orderByIndex < 0: orderBy = '' else: where, orderBy = where[:orderByIndex], where[orderByIndex:] return 'where deleted is null and (%s) %s' % (where, orderBy) else: return 'where deleted is null %s' % clauses ## Obj refs ## def fetchObjRef(self, objRef): """Fetch referenced object. Given an unarchived object reference, this method returns the actual object for it (or None if the reference is NULL or dangling). While this method assumes that obj refs are stored as 64-bit numbers containing the class id and object serial number, subclasses are certainly able to override that assumption by overriding this method. """ assert isinstance(objRef, long), 'type=%r, objRef=%r' % (type(objRef), objRef) if objRef == 0: return None else: klassId, serialNum = objRefSplit(objRef) if klassId == 0 or serialNum == 0: # invalid! we don't use 0 serial numbers return self.objRefZeroSerialNum(objRef) klass = self.klassForId(klassId) # Check if we already have this in memory first key = ObjectKey() key.initFromClassNameAndSerialNum(klass.name(), serialNum) obj = self._objects.get(key) if obj: return obj clauses = 'where %s=%d' % (klass.sqlSerialColumnName(), serialNum) objs = self.fetchObjectsOfClass(klass, clauses, isDeep=False) if len(objs) == 1: return objs[0] elif len(objs) > 1: # @@ 2000-11-22 ce: expand the msg with more information raise ValueError('Multiple objects.') else: return self.objRefDangles(objRef) def objRefInMem(self, objRef): """Return referenced object in memory. Returns the object corresponding to the given objref if and only if it has been loaded into memory. If the object has never been fetched from the database, None is returned. """ assert isinstance(objRef, long), 'type=%r, objRef=%r' % (type(objRef), objRef) if objRef == 0: return 0 else: klassId, serialNum = objRefSplit(objRef) if klassId == 0 or serialNum == 0: # invalid! we don't use 0 serial numbers return self.objRefZeroSerialNum(objRef) klass = self.klassForId(klassId) # return whether we have this object in memory key = ObjectKey() key.initFromClassNameAndSerialNum(klass.name(), serialNum) return self._objects.get(key) def objRefZeroSerialNum(self, objRef): """Raise serial number zero error. Invoked by fetchObjRef() if either the class or the object serial number is zero. """ raise ObjRefZeroSerialNumError(*objRefSplit(objRef)) def objRefDangles(self, objRef): """Raise dangling reference error. Invoked by fetchObjRef() if there is no possible target object for the given objRef. E.g., this can happen for a dangling reference. This method invokes self.warning() and includes the objRef as decimal, hexadecimal and class:obj numbers. """ raise ObjRefDanglesError(*objRefSplit(objRef)) ## Special Cases ## def filterDateTimeDelta(self, dtd): """Adapt DateTimeDeltas. Some databases have no TIME type and therefore will not return DateTimeDeltas. This utility method is overridden by subclasses as appropriate and invoked by the test suite. """ return dtd def sqlNowCall(self): """Get current DateTime.""" return 'CURRENT_TIMESTAMP' ## Self util ## def doneWithConnection(self, conn): """Invoked by self when a connection is no longer needed. The default behavior is to commit and close the connection. """ if conn is not None: # Starting with 1.2.0, MySQLdb disables autocommit by default, # as required by the DB-API standard (PEP-249). If you are using # InnoDB tables or some other type of transactional table type, # you'll need to do connection.commit() before closing the connection, # or else none of your changes will be written to the database. try: conn.commit() except Exception: pass conn.close() ## Debugging ## def dumpTables(self, out=None): if out is None: out = sys.stdout out.write('DUMPING TABLES\n') out.write('BEGIN\n') conn = None try: for klass in self.model().klasses().values(): out.write(klass.name() + '\n') conn, cur = self.executeSQL('select * from %s;' % klass.name(), conn) out.write(str(self._cursor.fetchall())) out.write('\n') finally: self.doneWithConnection(conn) out.write('END\n') def dumpKlassIds(self, out=None): if out is None: out = sys.stdout wr = out.write('DUMPING KLASS IDs\n') for klass in self.model().klasses().values(): out.write('%25s %2i\n' % (klass.name(), klass.id())) out.write('\n') def dumpObjectStore(self, out=None, progress=False): if out is None: out = sys.stdout for klass in self.model().klasses().values(): if progress: sys.stderr.write(".") out.write('%s objects\n' % (klass.name())) attrs = [attr for attr in klass.allAttrs() if attr.hasSQLColumn()] colNames = [attr.name() for attr in attrs] colNames.insert(0, klass.sqlSerialColumnName()) out.write(CSVJoiner.joinCSVFields(colNames) + "\n") # write out a line for each object in this class objlist = self.fetchObjectsOfClass(klass.name(), isDeep=False) for obj in objlist: fields = [] fields.append(str(obj.serialNum())) for attr in attrs: # jdh 2003-03-07: if the attribute is a dangling object reference, the value # will be None. This means that dangling references will _not_ be remembered # across dump/generate/create/insert procedures. method = getattr(obj, attr.pyGetName()) value = method() if value is None: fields.append('') elif isinstance(value, MiddleObject): fields.append('%s.%d' % (value.klass().name(), value.serialNum())) else: fields.append(str(value)) out.write(CSVJoiner.joinCSVFields(fields).replace('\r', '\\r')) out.write('\n') out.write('\n') out.write('\n') if progress: sys.stderr.write("\n")