v1.3.0 breaks block-call semantics of PG::Connection

[askreet]

Looking at the code, it seems like an oversight, but PG::Connection#initialize was previously implemented entirely in C, ending with this code:

if (rb_block_given_p()) {
    return rb_ensure(rb_yield, self, pgconn_finish, self);
}

In v1.3.0, the class function new is replaced with Ruby code here.

def new(*args, **kwargs)
	conn = self.connect_start(*args, **kwargs ) or
		raise(PG::Error, "Unable to create a new connection")

	raise(PG::ConnectionBad, conn.error_message) if conn.status == PG::CONNECTION_BAD

	conn.send(:async_connect_or_reset, :connect_poll)
end

The C implementation retains the semantics in connect_start (including a comment on 308 explaining it's role), but the block is not passed from new to connect_start in order to be called on line 326.

An example of some code that previously worked:

PG::Connection.open(admin_database_url) do |c|
  c.exec("DROP DATABASE #{database_name}")
end

With v1.3.0 and later, the inner block is not called.

[larskanis]

Yes, this is an oversight. I'll fix this.