Driver object should optionally contain connection information

[krlmlr]

Hannes suggests to store the information how to connect, i.e. the arguments to dbConnect(), in the driver object.

Solves a bunch of problems:

• reconnecting
• context for DBItest
• configuration of test databases

CC @hannesmuehleisen.

[jimhester]

Is there any concern this would increase the chances of exposing db passwords inadvertently? Obviously this is not really a concern in the testing cases.

[krlmlr]

We could do something like

drv <- RPostgres(..., password = function() get_password_from_keychain(...))

and have dbConnect() call the password getter upon connecting. Or partial specification:

drv <- RPostgres(...)
dbConnect(drv, password = ...)

[krlmlr]

• private .conn_args slot
• getter
• setter
• ensure arguments are picked up by dbConnect()

Maybe a new "DBIConnector" class, inheriting from "DBIDriver", could take care of forwarding to bare dbConnect(), to minimize downstream implementation effort?

[krlmlr]

Prefer aggregation over inheritance, they say.

[krlmlr]

Restrictions for arguments:

• must be named
• must not contain drv
• if function, must be callable without arguments

[krlmlr]

Pros of a new connector class:

• works immediately across all backends

Cons:

• it's possible to create a connector object with invalid argument names -- checking arguments still is the responsibility of dbConnect()
• if we want to support RSQLite::SQLite() returning a connector object instead of a driver, we need to tweak DBItest and perhaps implement dbDriver("DBIConnector")

[krlmlr]

Needs documentation tweaks.

[hannes]

Would be nice to have a close() method or something of the like

[krlmlr]

I forgot -- what do we need a close() method for?

[github-actions]

This old thread has been automatically locked. If you think you have found something related to this, please open a new issue and link to this old issue if necessary.