Re: Backend-internal SPI operations

Tom Lane wrote:
> "Mark Hollomon" <mhh@nortelnetworks.com> writes:
> > Just out of curiousity, is there technical reason there isn't
> > a (say) relisview attribute to pg_class?
>
> That might indeed be the most reasonable way to attack it, rather
> than having to go messing about looking for a matching rule.
> (Jan, any thoughts here?)
   The  right  way  IMHO would be to give views another relkind.   Then we could easily
   1.  detect if the final query after rewriting still tries  to       INSERT/UPDATE/DELETE  a  view  -  i.e.  "missing
rewrite       rule(s)".
 
   2.  disable things like LOCK etc.
   The problem here is, that the relkind  must  change  at  rule   creation/drop  time.  Fortunately rules on SELECT
aretotally   restricted to VIEW's since 6.4, and I don't see any reason to   change this.
 
   And  it's time to make more use of the relkind attribute. For   7.2, when we want to have tuple-set returns for
functions,we   might  want  to have structures as well (we talked about that   already, Tom). A structure is just a
row/typedescription.  A   function, returning a tuple or set of tuples, can return this   type or set of type as well
asany other existing  table/view   structure. So to create a function returning a set of tuples,   which have a
structuredifferent  from  any  existing  table,   someone   creates   a  named  structure,  then  the  function
returningtuples of that type.   These  structures  are  just   entries  in  pg_class, pg_attribute and pg_type.  There
isno   file or any rules, triggers etc. attached to them. They  just   describe a typle that can be built in memory.
 

> Adding a column to a core system table like pg_class is a good
> exercise for the student ;-) ... it's not exactly automated,
> and you have to find all the places that need to be updated.
> You might want to keep notes and prepare a writeup for the
> developer's FAQ.  I thought of that the last time I did something
> similar, but it was only at the end that I realized I should've
> been keeping notes to start with.
   Meetoo :-}


Jan

--

#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me.                                  #
#================================================== JanWieck@Yahoo.com #