Types work differently for test webservice and launchpadlib

[Jeroen T. Vermeulen]

I just did the following:
 * On an API-exported interface IFoo, export a first method bar().
 * Verify that Person and several other classes implement IFoo.
 * Test bar as webservice.named_get(person, 'bar'): succeeds.
 * Try to call person.bar in launchpadlib: no such method.

After a make clean / make build, the apidoc shows bar on a new class foo, not on person etc. So it seems that in effect webservice gives me more of a modern weak/dynamic type system than the real web-service API does.

It's probably not a good idea to have this succeed in testing but fail in the real world. The two should be more consistent.

[Gary Poster (gary)]

AFAIK, ideally we'd be testing with the "real" launchpadlib instance that's available. It's possible that it is still broken, but my last impression was that Leonard had addressed the issues bac had found. Then we can gradually get rid of the artificial webservice testing thing, unless someone clarifies why it is valuable separately.

[Jeroen T. Vermeulen (jtv)]

Using launchpadlib added about 25 seconds to my test, so I did not make that change.