> > > From: Tom Lane <t...@sss.pgh.pa.us> > > Hm, can you restore it into 9.2 either? > > AFAICS, pg_dump has absolutely no idea that it should be worried about the > value of xmloption, despite the fact that that setting affects what is > considered valid XML data. What's worse, even if it were attempting to do > something about xmloption, I don't see how it could deal with a case like > this where you have different values inside the same database that require > two different settings in order to parse. > > This isn't a 9.3.x bug, it's an aboriginal misdesign of the XML datatype. > Not sure what we can do about it at this point. Perhaps we could invent > a "document_or_content" setting that would tell xml_in to accept either > case? And then have pg_dump force that setting to be used during restore?
This sounds reasonable. My use case is purely as a document store, with the ability to perform xml parse functions against it – as such, I’m not concerned wether it’s a document or content – hence why we have both types recorded against that field. For the minute, I’m getting around the restore problem by mangling the dump such that the table is created using the text type rather than xml. This at least gets the data onto a 9.3 cluster, even if it’s cosmetically represented as text instead of xml. I can worry about the document vs content problem at a later stage. > > PS: BTW, I agree with the advice expressed by David J: under no > circumstances put any data you care about on 9.3.0. That release > was rather a disaster from a quality-control standpoint :-( > But that's unrelated to your XML issue. Ack. Thanks for the info. I’ll push the upgrade-path agenda a little harder.
