Compile Alembic's RenameTable the Postgres way

[bouk]

@graingert this fixes an issue where RenameTable would be compiled as

ALTER TABLE scheme."old" RENAME TO scheme."new"

[graingert]

Nice, but what did this compile as before? Edit: ah it's in your PR description

[bouk]

I put it in the description:

ALTER TABLE scheme."old" RENAME TO scheme."new"

Note the scheme that's included in the TO, which Postgres doesn't support

[graingert]

Do you know why this isn't pulled in by subclassing the PostgresalImpl?

[bouk]

It's because sqlalchemy uses the value in __dialect__, which RedshiftImpl overrides

This is where alembic defines the postgres override: https://github.com/zzzeek/alembic/blob/eb077a6c8d2c468736ba537eb1ab5400c9059aad/alembic/ddl/postgresql.py#L124

[graingert]

@zzzeek do you know the best way to inherit dialects' "compiles" declarations?

Eg in a way that if more are added I don't have to keep updating my subclass.

[zzzeek]

well Redshift should have an alembic impl called RedshiftImpl and use __dialect__ = 'redshift', subclassing PostgresqlImpl

[zzzeek]

so you have that....and....@compiles(RenameTable, 'redshift') should do the individual elements

[graingert]

Ah, so there's no nice way to automatically inherit the @compiles declarations

[zzzeek]

right so why not subclassing. well @compiles someday might want to take into account the hierarchy of SQLCompiler subclasses. not really sure. redshift is a very odd case.

[graingert]

Thanks for the PR, it's online at https://pypi.python.org/packages/3.4/s/sqlalchemy-redshift/sqlalchemy_redshift-0.1.2-py2.py3-none-any.whl#md5=9f6b52776b2f65695d62a3210fd46a57