GenericClient: Support COPY for Postgres

[seanlinsley]

I'd like to use GenericClient so pco_store's delete function can be optionally wrapped in a transaction, but pco_store currently relies on tokio-postgres's copy_in function, which isn't available through the generic client.

Would a PR be welcome to add copy_in support to GenericClient?

[seanlinsley]

It looks like tokio_postgres::GenericClient doesn't support copy_in so deadpool couldn't either. It seems like the function that relies on copy_in will have to accept a &deadpool_postgres::Client while others that could be wrapped in a transaction can accept &impl deadpool_postgres::GenericClient

[seanlinsley]

The problem is, I would like to DELETE from a table and then COPY in a transformed version of the data, all inside a transaction. It seems that tokio_postgres::GenericClient doesn't allow for that use case.

@sfackler thoughts on adding COPY support to GenericClient?

[seanlinsley]

Actually it seems that tokio-postgres does support transactions through SQL-standard BEGIN / COMMIT / ROLLBACK, so that would mean I don't need to use GenericClient. Is there some reason to prefer the transaction() function, other than its implicit rollback on Drop?

[bikeshedder]

As far as I understand @sfackler the only reason for Transaction borrowing mutably from Client is that you're forced to use it correctly. I remember discussing this with Steven a couple of years ago. I proposed to change the Transaction object to a Guard-only construct that doesn't contain the methods. Steven was highly against this. I didn't bother pushing on this topic.

As both Client and Transaction support the copy_in/copy_out methods. It is trivial to add this to the GenericClient. The GenericClient of deadpool-postgres is a 1:1 copy of the one from tokio-postgres.

I'll happily accept PRs to add missing functionality. Please also create a PR for tokio-postgres so feature parity is kept. Honestly I don't like that deadpool-postgres has its own GenericClient. The only reason for this is that tokio-postgres uses a sealed GenericClient trait that can't be implemented by other crates.

[seanlinsley]

Thanks for the feedback @bikeshedder. I'll hold off on opening PRs until we hear from @sfackler, for now just recommending that the transaction is managed with a macro: pganalyze/pco_store@cdb44c4