Tom Lane wrote:
> I dug through the archives and found that we've had this discussion
> before ;-). The basic argument for having the per-column form of
> pg_get_indexdef do what it does was that it's unreasonable for
> client-side code to try to disassemble an expression tree string,
> whereas extracting opclass info is a relatively straightforward
> exercise in joining. There are several past threads about this:
>
> http://archives.postgresql.org/pgsql-hackers/2003-07/msg00083.php
> http://archives.postgresql.org/pgsql-general/2005-11/msg01106.php
> http://archives.postgresql.org/pgsql-hackers/2006-06/msg00576.php
Ah, that takes me back :-)
> As of 8.3 that expands into also having to know the meaning of the
> bits in indoption[], which is kind of annoying, but not even close
> to being in the same league as reverse-compiling expressions.
Agreed.
> I think the current API expectation for pg_get_indexdef is that
> it produces only the index column/expression, and that we are
> very likely to break client-side code if we change that.
I'm sure there are far more catalog/API changes that'll affect any app
likely to be using this form of pg_get_indexdef, but I can live without
adding another.
> I don't have any objection to providing an additional new API that
> includes the opclass and ASC/DESC decoration in the output ... other
> than that I think it's a bit too late for 8.3; adding a function
> would mean forcing initdb, and I don't see any reasonable way to
> shoehorn two behaviors into the existing function signature.
Agreed - now is not the time to be adding new functions.
> Just out of curiosity, why is pgAdmin doing it this way at all?
> Seems it would be a lot easier to use the all-columns form of
> pg_get_indexdef than to cons up the display from fetches of each
> column individually.
We use the data in various UI elements as well as for reverse
engineering the SQL - it's easier to get it broken down than to parse it
back out of the complete definition.
/D
