SQLObjectStore.py

import sys

from MiddleObject import MiddleObject
from ObjectStore import ObjectStore, UnknownObjectError
from ObjectKey import ObjectKey
from MiddleKit.Core.ObjRefAttr import objRefJoin, objRefSplit
from MiscUtils import NoDefault, AbstractError, CSVJoiner
from MiscUtils import Funcs as funcs
from MiscUtils.DBPool import DBPool
from MiscUtils.MixIn import MixIn


class SQLObjectStoreError(Exception):
    """SQL object store error"""

class SQLObjectStoreThreadingError(SQLObjectStoreError):
    """SQL object store threading error"""

class ObjRefError(SQLObjectStoreError):
    """SQL object store object reference error"""

class ObjRefZeroSerialNumError(ObjRefError):
    """SQL object store serial number zero error"""

class ObjRefDanglesError(ObjRefError):
    """SQL object store object dangles error"""


aggressiveGC = False


class UnknownSerialNumberError(SQLObjectStoreError):
    """For internal use when archiving objects.

    Sometimes an obj ref cannot be immediately resolved on INSERT because
    the target has not yet been inserted and therefore, given a serial number.
    """

    def __init__(self, info):
        self.info = info

    def __repr__(self):
        return '%s: %s' % (self.__class__.__name__, self.info)

    def __str__(self):
        return str(self.info)


class UnknownSerialNumInfo(object):
    """For internal use when archiving objects.

    Attrs assigned externally are:
        sourceObject
        sourceAttr
        targetObject
    """

    def updateStmt(self):
        assert self.sourceObject.serialNum() != 0
        assert self.targetObject.serialNum() != 0
        sourceKlass = self.sourceObject._mk_klass
        assert sourceKlass
        sourceTableName = sourceKlass.sqlTableName()
        sourceSqlSerialName = sourceKlass.sqlSerialColumnName()
        return 'update %s set %s where %s=%s;' % (
            sourceTableName, self.sourceAttr.sqlUpdateExpr(self.targetObject),
            sourceSqlSerialName, self.sourceObject.serialNum())

    def __repr__(self):
        s = []
        for item in self.__dict__.items():
            s.append('%s=%r' % item)
        s = ' '.join(s)
        return '<%s %s>' % (self.__class__.__name__, s)


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().
        """
        if clausesArgs:
            cur.execute(sql, clausesArgs)
        else:
            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 sorted(self.model().klasses().values(), key=lambda k: k.name()):
            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 Model(object):

    def sqlDatabaseName(self):
        """Return the name of the database.

        This is either the 'Database' setting or self.name().
        """
        name = self.setting('Database', None)
        if name is None:
            name = self.name()
        return name


class MiddleObjectMixIn(object):

    def sqlObjRef(self):
        """Return the 64-bit integer value that refers to self in a SQL database.

        This only makes sense if the UseBigIntObjRefColumns setting is True.
        """
        return objRefJoin(self.klass().id(), self.serialNum())

    def sqlInsertStmt(self, unknowns, id=None):
        """Return SQL insert statements.

        Returns the SQL insert statements for MySQL (as a tuple) in the form:
            insert into table (name, ...) values (value, ...);

        May add an info object to the unknowns list for obj references that
        are not yet resolved.
        """
        klass = self.klass()
        insertSQLStart, sqlAttrs = klass.insertSQLStart(includeSerialColumn=id)
        values = []
        append = values.append
        extend = values.extend
        if id is not None:
            append(str(id))
        for attr in sqlAttrs:
            try:
                value = attr.sqlValue(self.valueForAttr(attr))
            except UnknownSerialNumberError as exc:
                exc.info.sourceObject = self
                unknowns.append(exc.info)
                if self.store().model().setting('UseBigIntObjRefColumns', False):
                    value = 'NULL'
                else:
                    value = ('NULL', 'NULL')
            if isinstance(value, basestring):
                append(value)
            else:
                # value could be sequence for attrs requiring multiple SQL columns
                extend(value)
        if not values:
            values = ['0']
        values = ','.join(values)
        return insertSQLStart + values + ');'

    def sqlUpdateStmt(self):
        """Return SQL update statement.

        Returns the SQL update statement of the form:
            update table set name=value, ... where idName=idValue;
        Installed as a method of MiddleObject.
        """
        assert self._mk_changedAttrs
        klass = self.klass()
        res = []
        for attr in self._mk_changedAttrs.values():
            res.append(attr.sqlUpdateExpr(self.valueForAttr(attr)))
        res = ','.join(res)
        res = ('update ', klass.sqlTableName(), ' set ', res, ' where ',
            klass.sqlSerialColumnName(), '=', str(self.serialNum()))
        return ''.join(res)

    def sqlDeleteStmt(self):
        """Return SQL delete statement.

        Returns the SQL delete statement for MySQL of the form:
            delete from table where idName=idValue;
        Or if deletion is being marked with a timestamp:
            update table set deleted=Now();
        Installed as a method of MiddleObject.
        """
        klass = self.klass()
        assert klass is not None
        if self.store().model().setting('DeleteBehavior', 'delete') == 'mark':
            return 'update %s set deleted=%s where %s=%d;' % (
                klass.sqlTableName(), self.store().sqlNowCall(),
                klass.sqlSerialColumnName(), self.serialNum())
        else:
            return 'delete from %s where %s=%d;' % (klass.sqlTableName(),
                klass.sqlSerialColumnName(), self.serialNum())

    def referencingObjectsAndAttrsFetchKeywordArgs(self, backObjRefAttr):
        if self.store().setting('UseBigIntObjRefColumns'):
            return dict(refreshAttrs=True, clauses='WHERE %s=%s'
                % (backObjRefAttr.sqlColumnName(), self.sqlObjRef()))
        else:
            classIdName, objIdName = backObjRefAttr.sqlColumnNames()
            return dict(refreshAttrs=True, clauses='WHERE (%s=%s AND %s=%s)'
                % (classIdName, self.klass().id(), objIdName, self.serialNum()))

MixIn(MiddleObject, MiddleObjectMixIn)
    # Normally we don't have to invoke MixIn()--it's done automatically.
    # However, that only works when augmenting MiddleKit.Core classes
    # (MiddleObject belongs to MiddleKit.Run).


import MiddleKit.Design.KlassSQLSerialColumnName


class Klass(object):

    _fetchSQLStart = None  # help out the caching mechanism in fetchSQLStart()
    _insertSQLStart = None  # help out the caching mechanism in insertSQLStart()

    def sqlTableName(self):
        """Return the name of the SQL table for this class.

        Returns self.name().
        Subclasses may wish to override to provide special quoting that
        prevents name collisions between table names and reserved words.
        """
        return self.name()

    def fetchSQLStart(self):
        if self._fetchSQLStart is None:
            attrs = self.allDataAttrs()
            attrs = [attr for attr in attrs if attr.hasSQLColumn()]
            colNames = [self.sqlSerialColumnName()]
            colNames.extend([attr.sqlColumnName() for attr in attrs])
            self._fetchSQLStart = 'select %s from %s ' % (','.join(colNames), self.sqlTableName())
        return self._fetchSQLStart

    def insertSQLStart(self, includeSerialColumn=False):
        """Return a tuple of insertSQLStart (a string) and sqlAttrs (a list)."""
        if self._insertSQLStart is None:
            res = ['insert into %s (' % self.sqlTableName()]
            attrs = self.allDataAttrs()
            attrs = [attr for attr in attrs if attr.hasSQLColumn()]
            fieldNames = [attr.sqlColumnName() for attr in attrs]
            if includeSerialColumn or len(fieldNames) == 0:
                fieldNames.insert(0, self.sqlSerialColumnName())
            res.append(','.join(fieldNames))
            res.append(') values (')
            self._insertSQLStart = ''.join(res)
            self._sqlAttrs = attrs
        return self._insertSQLStart, self._sqlAttrs


class Attr(object):

    def shouldRegisterChanges(self):
        """Return self.hasSQLColumn().

        This only makes sense since there would be no point in registering
        changes on an attribute with no corresponding SQL column. The standard
        example of such an attribute is "list".
        """
        return self.hasSQLColumn()

    def hasSQLColumn(self):
        """Check if there is a correlating SQL column.

        Returns true if the attribute has a direct correlating SQL column in
        its class' SQL table definition.
        Most attributes do. Those of type list do not.
        """
        return not self.get('isDerived', False)

    def sqlColumnName(self):
        """Return the SQL column name corresponding to this attribute-

        This is consisting of self.name() + self.sqlTypeSuffix().
        """
        if not self._sqlColumnName:
            self._sqlColumnName = self.name()
        return self._sqlColumnName

    def sqlValue(self, value):
        """Return SQL for Python value.

        For a given Python value, this returns the correct string for use in a
        SQL statement. Subclasses will typically *not* override this method,
        but instead, sqlForNonNone() and on rare occasions, sqlForNone().
        """
        if value is None:
            return self.sqlForNone()
        else:
            return self.sqlForNonNone(value)

    def sqlForNone(self):
        return 'NULL'

    def sqlForNonNone(self, value):
        return repr(value)

    def sqlUpdateExpr(self, value):
        """Return update assignments.

        Returns the assignment portion of an UPDATE statement such as:
            "foo='bar'"
        Using sqlColumnName() and sqlValue(). Subclasses only need to
        override this if they have a special need (such as multiple columns,
        see ObjRefAttr).
        """
        colName = self.sqlColumnName()
        return colName + '=' + self.sqlValue(value)

    def readStoreDataRow(self, obj, row, i):
        """By default, an attr reads one data value out of the row."""
        value = row[i]
        obj.setValueForAttr(self, value)
        return i + 1


class BasicTypeAttr(object):
    pass


class IntAttr(object):

    def sqlForNonNone(self, value):
        return str(value)
        # it's important to use str() since an int might point
        # to a long (whose repr() would be suffixed with an 'L')


class LongAttr(object):

    def sqlForNonNone(self, value):
        return str(value)


class DecimalAttr(object):

    def sqlForNonNone(self, value):
        return str(value)  # repr() will give Decimal("3.4")


class BoolAttr(object):