Yup. This is the silly thing where, in interactive mode, erlang will such current working for for modules. Since we don't "cd" in the script to a guaranteed readable dir, spurious errors abound. A proper OTP launch would not do this as all paths are defined up front. Sent from the pool side with a glass of wine in hand. On 3 Oct 2012, at 17:00, Adam Kocoloski wrote: > On Oct 3, 2012, at 10:54 AM, Dave Cottlehuber wrote: > >> On 3 October 2012 16:49, Adam Kocoloski wrote: >>> Yep, that matches up with my experience. >>> >>> Adam >>> >>> On Oct 3, 2012, at 10:47 AM, Martin Hewitt wrote: >>> >>>> This one's further complicated by the fact that, after spitting out errors at start, CouchDB appears to operate normally with no further errors. Generating a view produces another error, but the view is generated correctly, and further view updates do not produce errors. >>>> >>>> Very curious behaviour. >>>> >>>> Martin >> >> Aaah this vaguely comes back to me now, hopefully enough to trigger >> somebody else's memory with something correct. >> >> I think it's due to the search path for the erlang code loader >> including directories it can't check/view. So if you can do a dtrace >> or truss or whatever , this should turn up pretty quickly. >> >> A+ >> Dave > > Oh, that makes a ton of sense. Thanks for the tip!