Roger Leigh wrote:
My current impression is that the mess was caused by creating the wanna-build database with debversion not installed resulting in the creation of an incomplete debversion type (the wanna-build scripts contain a literal copy of the sql for setting up a debversion type rather than referencing it, this seems like a flaw to me because it means that if what needs to be done changes the wanna-build scripts won't pick it up) .Thanks for the hint, changing the column type from debversion to varchar and back to debversion appears to have fixed things (at least I was able to get a log that failed to inject before to inject succesfully, lets hope this fix works in general). I had to delete some views before postgres would let me mess with the column type (and recreate them after) which was a bit of a pain but not too big a deal.That's good to know, because this was just a guess! You might also potentially want to recreate any indexes on those columns as well to ensure that you're using the debversion index operators (it's possible this happens automatically--I'm not entirely certain of the postgres internal behaviour here). Do you know what triggered this condition? I've never seen this type of behaviour before, so it would be good to know what the cause was for the future.
But I really don't know for sure, all I know is I was getting an obtuse error message (and only getting it with some data not all) from a peice of software I rarely tangle with and which noone I asked could provide a methodical way to debug, just various variants on "try this". It seems to be working now but I don't really know how or why presumablly changing the column type and changing it back reset some stuff but I dunno what exactly.
Now I understand why people who are rich enough to do so hire DBAs.