On Tue, Nov 27, 2012 at 01:59:04AM -0800, Jeff Davis wrote:
> On Wed, 2012-11-21 at 15:27 +0000, Simon Riggs wrote:
> > It would be useful if we issued a NOTICE when an ambiguity is
> > introduced, rather than when using it.
> >
> > Like Bison's reporting of reduce conflicts.
>
> This brings up a very important point, which is that a lot of the code
> is frozen in applications yet invisible at DDL time. So we have to be
> careful that DDL changes have a reasonable impact on the ability to
> continue to compile and execute the previously-working SQL received from
> the applications.
>
> In other words, as I said in another reply, we want to avoid cases where
> something seemingly innocuous (like creating a function) causes
> previously-working SQL to fail due to ambiguity.
>
> As Tom said, detecting the ambiguity at DDL time is not easy, so I'm not
> suggesting that. And I know that creating a function can already cause
> previously-working SQL to fail. I'm just saying we should be careful of
> these situations and not make them more likely than necessary.
For me this highlights why looking at how application languages handle
overloading might not be as relevant --- most language don't have
possible-conflicting functions being created at run-time like a database
does. The parallels in how other databases treat overloading is
relevant.
-- Bruce Momjian <bruce@momjian.us> http://momjian.us EnterpriseDB
http://enterprisedb.com
+ It's impossible for everything to be true. +
