Re: Bug in pg_describe_object (was: Re: [HACKERS] obj_unique_identifier(oid))

2011/1/10 Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>:
> There was never any intention that that code produce a guaranteed-unique
> identifier; it's only meant to be a humanly useful identifer, and this
> patch seems to me to mostly add noise.

For all objects, except for these pg_amproc regclass, the function
does already generate unique strings. They are guaranteed to be unique
thanks to every component of the unique constraints in alll pg_*
tables are included in the unique text identifier.

It makes a lot more sense to fix the function to return a unique
string also for pg_amproc, than to introduce a entirely new function
which returns a unique string identifier. It would hardly break
anything and I think you exaggerate the noise factor.

I can think of numerous reasons why it is absolutely necessary to
provide a function generating unique identifiers for objects:

a) To allow comparing all objects in two different databases, by
comparing objects with the same identifier. This cannot be done using
the oids, since they naturally differ between databases.

b) To draw nice human readable digraphs in the .dot format , instead
of drawing relations digraphs of classid.objid.subobjid.

c) OIDs are probably misused in a lot of applications, due to
misunderstandings of what they are and not are, I for one didn't know
they are not necessarily unique, but only within their regclass. It
would be better to encourage users to use a text string if they need
to refer to a unique objects in their application, than to force them
to use OIDs (or in combination with the regclass, almost as bad), in
lack of something better.
While you could build your own query to generate a unique string,
based on all the columns defining the unique constraint for each
class, doing so is very cumbersome and requires a lot of
postgres-guru-knowledge.

I think it would be a big improvement and increase the number of
possible use cases of the existing pg_describe_object function if the
documentation would say "the returned text is guaranteed to be unique
for each object".

--
Best regards,

Joel Jacobson
Glue Finance

E: jj(at)gluefinance(dot)com
T: +46 70 360 38 01

Postal address:
Glue Finance AB
Box  549
114 11  Stockholm
Sweden

Visiting address:
Glue Finance AB
Birger Jarlsgatan 14
114 34 Stockholm
Sweden