feat(views): Add schema support for views & MVs

[kozlek]

This PR mainly adds schema support for create_view(), create_materialized_view() and refresh_materialized_view().

Also, it adds a populate keyword argument to create_materialized_view(). When set to False, the materialized view will be created with the WITH NO DATA flag, which is useful when migrating a database schema with heavy materialized views.

Regarding cascade_on_drop when creating views, it is now available for create_materialized_view().
However, it has been set to None by default, instead of the previous True, in order to better match PostgreSQL default's mode of operation.
I understand this is a breaking change, and I'll be happy to revert that part if you're not confortable with that.

Others remarks about the PR:

• it introduces the usage of f-string, which is supported only from Python 3.6 but it seems to match the actual requirements for the project
• it does not include typing annotations as the project currently doesn't have any. If you want me to add them, I can do it.
• it set some arguments like schema and populate as keyword-only arguments, as I believe they are non trivial arguments. It can be removed if you don't like that.

See also: #427

[kurtmckee]

It's fine to use f-strings. Type annotations will be helpful because I'd like to see strong type annotations in the future.

The tests are currently failing; please review the CI logs for details. The tests also appear to be incomplete because several code paths are not exercised. In particular, cascade and populate can be one of three values now, and I don't think all of the possibilities are getting tested.

I like that cascade_on_drop is None to better match postgres defaults, but this is a breaking change. I cannot merge this because I'm not familiar with views or how sqlalchemy-utils deals with breaking changes. @kvesteri, would you provide feedback about this change, including (a) converting parameters to be keyword-only, (b) changing the cascade_on_drop behavior, (c) anything else you see?

[kozlek]

Thanks for your review @kurtmckee.

Pipeline should be fixed by now, and I added typing annotations to the whole views.py.

Regarding missing tests related to cascade option, I realised the actual cascade test suite (based on life_cycle method) is a total non-sense.

class TrivialViewTestCases:
    def life_cycle(
        self,
        engine,
        metadata,
        column,
        cascade_on_drop
    ):
        __table__ = create_view(
            name='trivial_view',
            selectable=sa.select(*_select_args(column)),
            metadata=metadata,
            cascade_on_drop=cascade_on_drop
        )
        __table__.create(engine)
        __table__.drop(engine)

create_view is returning a Table object, while views DDL operations (CreateView / DropView )are added as events to be triggered at metadata creation / deletion.
That's why when we call __table__.create(engine), we're calling a Table DDL operation leading to the creation of a table instead of a view.
Postgres logs as a proof:

CREATE TABLE trivial_view (
                id SERIAL NOT NULL,
                PRIMARY KEY (id)
        );
DROP TABLE trivial_view;

We're not testing anything here, at least not something related to views.

In order to complete this PR, I'd like to add missing tests as you asked, but I'd also like to rewrite the whole cascade related tests suite if that's okay for you.

[halali]

any update in this ?

[disbr007]

I would love to see this feature merged.