It was a fun effect watching it when I was testing. Hrvoje was right, these things are too heavy-weight to be trusted to be closed by garbage collection. I've been learning programming style as I go along. It's been fifteen years since I've done any database programming at this level and I'm still learning how to do it in Emacs Lisp. :-) My latest code has everything wrapped in (unwind-protect <> (pq-finish)) wrappers.

I will update all the examples in the manual before there is any 21.2 release. Most people when learning how to program a new library/language will start with any provided examples and adjust them to do what they want. It is highly important that the provided examples are good ones.

This is a real interface instead of a hacked up one parsing textual output of a non-extensible application. Hacking the output of closed programs is sometimes the only reasonable way to deal with them, however, we are gradually taking over the world and open source and open interfaces are becoming a lot more common. I greatly prefer using library-style precise interfaces over ad-hoc parse-the-output interfaces because dealing with libraries is inherently more reliable. Or put another way, I have many reasons why I want this interface to be successful in the short-term even if it eventually gets surpassed by an RDBMS independent one.