Re: Fixing pg_dump

Christopher Kings-Lynne <chriskl@familyhealth.com.au> writes:
> I tried adding the extra scan in and it as all well and good up until 
> the second where I realised that the TocEntry struct has no field that 
> allows me to know the correct way of finding the full descriptor of each 
> object.

Ugh.  Definitely an oversight.  Don't suppose you want to think about
pulling the name out of the DROP command ;-) ?

> I'm running out of time unfortunately, and I need to know from you 
> whether I should go back to my work on making owner and acl TOC entries 
> fully independent?  All this means is that people restoring pre-7.5 
> binary dumps into 7.5 will not get the owner fixes...  But people using 
> the binary format to upgrade seems like a pretty rare case to me!

Hardly --- for instance, people using large objects have no other
choice.

You can *not* change pg_restore in a way that will make it impossible
for such people to restore their dumps (and no, I don't think it will
fly to tell someone after the fact they should have used 7.5 pg_dump...)

Maybe it's sufficient to have a backwards-compatibility mode in which
the SET SESSION AUTH commands still get issued same-as-ever.  In fact,
you could just automatically do that if you see the archive version is
too old to have ALTER OWNER support.

On the whole though, I think editing the DROP commands might be the best
way.  Are there any cases where that would actually not work?
        regards, tom lane