Thread: types oids

Hello,

I'm programming with pqlib from Pascal (FPC) and trying to build some
classes to encapsulate result sets in a similar way as Delphi.

I need to know the type of every field and I've found that PQftype returns
an OID. OK, I can query pg_type system table and find out type information.

But this leads me to new questions because it seems that standard types
aren't hardcoded anywhere. pg_type tells me the name, size and some other
information about the type. But what about the "kind" of the type (numeric,
text, date...)?

I can only think of comparing the name with a list of names for standard
types. Is there another simpler method? A standard relation between standard
types and their OIDs?

If the answer to last question is "no", can I really trust at least the name?
If so, since I don't want to make the metadata query before each "regular"
query, I'll have to cache it. Is it OK to do it once per connection?

I can even think that.... is the user (admin) allowed to change the names
of standard types?

Another, more trivial question: do you know where to find nice logos
for postgres?

Thanks in advance

 Nico


http://www.clubdelphi.com/nico/

Nicolas Aragon <nyco@jet.es> writes:
> But this leads me to new questions because it seems that standard types
> aren't hardcoded anywhere. pg_type tells me the name, size and some other
> information about the type. But what about the "kind" of the type (numeric,
> text, date...)?

You pretty much have to know from the name what the datatype is.
Postgres doesn't have any type categorization like that.  (Maybe it
should ... I think there are some hacks in the parser that amount to
categorizing the standard datatypes, but that knowledge is not visible
outside the parser.)

> I can only think of comparing the name with a list of names for standard
> types. Is there another simpler method? A standard relation between standard
> types and their OIDs?

In reality, the OIDs assigned to standard types like "bool" and "int4"
are unlikely ever to change; they certainly will not change faster than
once per Postgres major release.  You could probably get away with
hard-wiring the knowledge that OID 16 == bool, say.  If that seems a
little too ugly, reading the OIDs for known types during startup is
plenty good enough.

If you are dealing with user-defined types, looking up the OIDs at
startup is the way to go.  (User type OIDs could change across a
database dump/reload, so hardwiring those would be a bad idea.)

> I can even think that.... is the user (admin) allowed to change the names
> of standard types?

Not if he wants things to keep working ;-)

> Another, more trivial question: do you know where to find nice logos
> for postgres?

See the website.
        regards, tom lane