include_dir catch-22

There is, IMO, a significant oversight with the include_dir feature.

If a distributor wants to enable it by default at initdb time, they
can't just turn it on in postgresql.conf.sample, because initdb will die
when the postgres backend refuses to start because the configdir is missing.

Yet the configdir cannot exist, or initdb will refuse to run.

IMO we should just treat the configdir as implicitly if_exists, but
otherwise, there needs to be a separate if_exists option.

I'll gladly submit a patch to fix this for 9.4, I just want an opinion
on which way to go first.

("initdb then 'sed' the config file afterwards" is not an answer)

--
Craig Ringer http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services