Re: Name column

On Fri, Sep 24, 2010 at 11:43 AM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> =?iso-8859-1?B?QW5kcukgRmVybmFuZGVz?= <andre(dot)de(dot)camargo(dot)fernandes(at)hotmail(dot)com> writes:
>>> On Fri, Sep 24, 2010 at 6:35 AM, Heikki Linnakangas
>>> <heikki(dot)linnakangas(at)enterprisedb(dot)com> wrote:
>>> I'm starting to wonder if we should think about deprecating this
>>> behavior.  It is awfully confusing and unintuitive.
>
>> I agree, it is very unintuitive.
>> +1  for deprecating this behavior.
>
> -1.  There's nothing wrong with the function-as-a-computed-column
> feature, and it seems likely that taking it away will break applications.
>
> What we are getting bit by is that I/O coercions to string types can be
> specified this way.  Maybe what we ought to do is remove just that one
> capability.  It'd be a bit non-orthogonal, but seems fairly unlikely to
> break anything, especially since we only began to allow such things
> recently (in 8.4 looks like).

I think that might be an improvement, but I'm not convinced it goes
far enough. What evidence do we have that anyone is relying on this
behavior in applications? Every report I've heard of it involved
someone being surprised that it worked that way.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise Postgres Company