Re: pg_dump doesn't dump binary compatible casts

Tom Lane wrote:

> Jan Wieck <JanWieck@Yahoo.com> writes:
>> I classify this problem as a bug. Objections?
> 
> The question is not whether it is a bug, the question is what is correct
> behavior instead.

Not yet, but when we have a decision and a fix it'll be the criterium 
for applying it to 7.4 and maybe backpatch into 7.3.

> 
>> IMHO a binary compatible cast should be dumped if one or both namespaces 
>> of the underlying data types is included in the dump.
> 
> That would just replace one bug with another (viz, failure to restore
> if one of the datatypes wasn't included in the dump).  Also, what of
> user-defined casts between two system datatypes?  pg_catalog is never
> considered part of the dump AFAIR.
> 
> The first idea that came to mind is to dump the cast if both underlying
> types are either system types or included in the dump.  But I think that
> would end up dumping built-in binary casts, which is no good either.
> I hate to think of looking at the cast's OID vs. lastsysoid to decide
> if it was a built-in cast, but maybe there's no other way.

User defined casts between builtin types are a bad thing anyway. They 
are only good for special backward compatibility cases or to support 
applications that are broken by design anyway. And a user defined binary 
cast between builtin types is just the worst case. Thus, braking the 
application with every dump+restore to remind the programmer that 
there's still something he has to fix is the _right thing_.

CREATE CAST for that beast should spit out a popup asking
    "Do you really ..."

with "No" being the default! Well, a WARNING might do for now.

That eliminated, a cast should be dumped if one or more of the three 
objects (source, target and function) are not builtin AND all the 
non-builtin objects belong to namespaces included in the dump.


Jan

-- 
#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me.                                  #
#================================================== JanWieck@Yahoo.com #