Anonymous Classes

[Lachlan Gunn]

Hello.

I have a library (SQLObject) that stores data as class variables. I would
like to set a class variable in a certain context, specific to a certain
instance of an object. This would require some sort of anonymous class. I
have attempted to use the following code to set the connection string:

| class SQLStorage:
| def __init__(self, c, debug = False):
| config = StorageConfiguration(c)
|
| connection = sqlobject.connectionForURI(config.databaseString)
| if debug:
| connection.debug = True
|
| # I don't know whether this is right. My belief is that we can
| # subclass each table and use _connection on them.
| class newDatum(DatumTable):
| _connection = connection
|
| class newMetadatum(MetadatumTable):
| _connection = connection
|
| class newChecksum(ChecksumTable):
| _connection = connection
|
| self.__SQLDatum = newDatum
| self.__SQLMetadatum = newMetadatum
| self.__SQLChecksum = newChecksum

This does not work; Python complains that the classes already exist when
SQLObject is instantiated for a second time. This has led me to try
instantiating a subclass using DatumTable.__class__ and friends, but this
requires setting a class name, which, again, limits me to a single
instance. I *could* have a counter that appends a number to each class
name, but that's a fairly ugly solution; is it possible to create an
anonymous class in Python?

Thanks,
Lachlan.

[Bruno Desthuilliers]

Lachlan Gunn a écrit :

Hello.

I have a library (SQLObject) that stores data as class variables. I would
like to set a class variable in a certain context, specific to a certain
instance of an object. This would require some sort of anonymous class. I
have attempted to use the following code to set the connection string:

| class SQLStorage:

You probably want:

class SQLStorage(object):

| def __init__(self, c, debug = False):
| config = StorageConfiguration(c)
|
| connection = sqlobject.connectionForURI(config.databaseString)
| if debug:
| connection.debug = True
|
| # I don't know whether this is right. My belief is that we can
| # subclass each table and use _connection on them.
| class newDatum(DatumTable):
| _connection = connection
|
| class newMetadatum(MetadatumTable):
| _connection = connection
|
| class newChecksum(ChecksumTable):
| _connection = connection
|
| self.__SQLDatum = newDatum
| self.__SQLMetadatum = newMetadatum
| self.__SQLChecksum = newChecksum

Are you sure you need name mangling here ?

This does not work; Python complains that the classes already exist when
SQLObject is instantiated for a second time. This has led me to try
instantiating a subclass using DatumTable.__class__ and friends, but this
requires setting a class name, which, again, limits me to a single
instance. I *could* have a counter that appends a number to each class
name, but that's a fairly ugly solution; is it possible to create an
anonymous class in Python?

Yes. Using type() :

Help on class type in module __builtin__:

class type(object)
| type(object) -the object's type
| type(name, bases, dict) -a new type

>>class Toto(object):

.... pass
....

>>class Tutu(object):

.... def __init__(self, what):
.... self._toto = type('Autototo', (Toto,), dict(_what=what))
....

>>t = Tutu(42)
t._toto

<class '__main__.Autototo'>

>>t._toto._what

42

>>t._toto()

<__main__.Autototo object at 0xb7c3806c>

>>t2 = Tutu(1138)
t2._toto

<class '__main__.Autototo'>

>>t2._toto._what

1138

[Lachlan Gunn]

Thanks for your help.

After trying your suggestion for a while, I eventually realised that
SQLObject was doing something funny in its metaclass that was causing the
issue, which I got around by mangling the class names with a counter.