Alright, I've totally got a prototype working for creating datasources (V10) as well. It's ugly but wanted to get it out for discussion.

    @classmethod
    def from_scratch(cls, caption, connections):
        from uuid import uuid4
        unique = lambda x: x + '.' + str(uuid4()).replace('-', '')[:28]
        root = ET.Element('datasource')
        root.set('caption', caption)
        root.set('version', '10.0')
        outer_connection = ET.SubElement(root, 'connection')
        outer_connection.set('class', 'federated')
        named_conns = ET.SubElement(outer_connection, 'named-connections')
        for conn in connections:
            nc = ET.SubElement(named_conns, 'named-connection')
            nc.set('name', unique(conn.dbclass))
            nc.append(conn._connectionXML)
        print(ET.tostring(root))
        return cls(root)

I don't actually know the rules for what makes a good 'name' -- it appears to be dbclass + a 28 character random string ('postgres.1of3kl00aoax5d1a1ejma1397430'), it's definitely not a UUID though

Alright, pushed my attempt.

@RussTheAerialist I'm not sure if it's overcomplicating things or not, but it feels like this logic could live in an, bleck, DatasourceBuilder, or something like that. It could take in a version parameter and know how to build 9.X vs 10.X as well.

Of course, none of that addresses the problem that these are 'bare' datasources without relations, metadata-records, or fields.

Thoughts?

🚀 this will need to be changed once I get all of the editor+physical+logical splitting done but for now, yay!

Awesome! At least it'll let us play with some examples, and the tests will stay valid.