Polymorphism for schema

[fangpenlin]

We have a scenario, say you have a message schema

class MessageSchema(colander.MappingSchema):
    header = EventHeader()
    payload = Payload()

For the payload, you can define it like this

class Payload(colander.MappingSchema):
    type = colander.SchemaNode(colander.String())
    # and other attributes ...

It works fine if you can define all possible fields to the payload schema, however, in many cases, you simply can't. A better way is to define subclasses. Here we let Log inherits Payload

class Log(Payload):
    message = colander.SchemaNode(colander.String())
    severity = colander.SchemaNode(colander.String(), validators=[
        colander.OneOf((
            'debug',
            'info',
            'warning',
            'error',
        ))
    ])

By doing that, when you do serializing or deserializing, the payload = Payload() is still there in MessageSchema, so there is no way it can know which subclass schema to use. To solve that problem, we will need to have something like the polymorphism mechanism like the one from SQLAlchemy:

class MyClass(Base):
    __tablename__ = 'my_table'
    id = Column(Integer, primary_key=True)
    type = Column(String(50))
    alt = Column("some_alt", Integer)

    __mapper_args__ = {
        'polymorphic_on' : type
    }

http://docs.sqlalchemy.org/en/rel_0_9/orm/mapper_config.html#class-mapping-api

For colander, that could be something like

class Payload(AbstractSchema):
    type = colander.SchemaNode(colander.String())
    __mapper_args__ = {
        'polymorphic_on': 'type',
    }

class Log(Payload):
    type = 'log'
    message = colander.SchemaNode(colander.String())
    severity = colander.SchemaNode(colander.String(), validators=[
        colander.OneOf((
            'debug',
            'info',
            'warning',
            'error',
        ))
    ])

I will submit a pull request for this soon

[fangpenlin]

what do you guys think about this?

[jaseemabid]

@victorlin Can you provide a failing test case to solidify the idea?

[tisdall]

I really like this idea. I'm working on implementing a standard using JSON that does this very thing. It has one attribute that defines the type and the rest of the attributes can be validated depending on that "type" value.

[antolantic]

Great job @fangpenlin , I am surprised that the fix #192 has not been pull requested ?
It's a really needed feature for colander, any chance to merge it into trunk ?

[tseaver]

@antolantic there is no PR yet for the work @fangpenlin has done on her fork.

[fangpenlin]

@tseaver @antolantic Hi guys, I am not currently working on any project that's based on colander, and I don't have time to work on it now. You're very welcome to continue my previous work.

[antolantic]

Should I fork the @fangpenlin repo and make a PR then ?

[mmerickel]

An administrative note: At the very least all of the commit authors need to have put their names into contributors.txt before anyone will consider merging such a PR into the official colander repo.

[fangpenlin]

@mmerickel I've added my name to contributors on top of my original branch

fangpenlin@3a01216

does it help?

[antolantic]

@fangpenlin : could you initiate the PR ?