Skip site navigation (1) Skip section navigation (2)

Re: Cannot assign ROWTYPE, RECORD variables in PL/pgSQL

From: elein <elein(at)sbcglobal(dot)net>
To: Neil Conway <neilc(at)samurai(dot)com>, josh(at)agliodbs(dot)com
Cc: <pgsql-bugs(at)postgresql(dot)org>, elein(at)norcov(dot)com
Subject: Re: Cannot assign ROWTYPE, RECORD variables in PL/pgSQL
Date: 2002-11-08 03:11:29
Message-ID: (view raw, whole thread or download thread mbox)
Lists: pgsql-bugspgsql-sql
Deep copy with User defined data types sometimes get a little 
wild, possibly with alignment and memory context.  For example 
a UDT which is a char followed by an int might be tricky to recognize
that alignment might be needed.  It might even be better to  have 
the UDT writer write their own deep copy function if their type 
is not compatible with a straight memcpy.

One of the other reasons this was a real PITB at informix was that 
columns could also contain row (composite) types.  We do not 
have that feature (yet?), but if deep copy is done in a type 
blind way which is open to adding recursion we would not shut 
the door on the possibility.  Tables have rows which have columns
containing rows which have columns containing udts and rows....

However, I suspect that postgresql row handling is a lot cleaner than
the informix row handling (with or without rows as columns) and it 
has been a while since I looked at the problem so maybe it is a 
non-issue.  But I'm raising it just in case...


On Thursday 07 November 2002 13:56, Neil Conway wrote:
> Josh Berkus <josh(at)agliodbs(dot)com> writes:
> > create or replace function rowtype_test ()
> > returns text as '
> > declare this_row candidates%rowtype;
> > 	that_row candidates%rowtype;
> > begin
> > select * into this_row
> > from candidates;
> >
> > that_row := this_row;
> >
> > return that_row.first_name;
> >
> > end;'
> > language 'plpgsql';
> > =======================================
> >
> > ... it will error out at the assignment "that_row := this_row".
> So we'd want a deep copy, right?
> > The only way to populate that_row with a copy of this_row is by
> > re-querying the source table.
> Well, you can also iterate through the fields of this_row and assign
> them to that_row manually -- of course, that's not much better.
> > While a relatively easy workaround, this behaviour is annoying and
> > inconsistent.  It would be nice to fix in 7.3.1 or 7.4.
> Unless anyone sees a problem with this, I'll work on this. I
> definately think it's inappropriate for 7.3.1 though.
> Cheers,
> Neil

In response to

pgsql-bugs by date

Next:From: Mats LofkvistDate: 2002-11-08 15:21:14
Subject: JDBC access broken in 7.3 beta
Previous:From: Josh BerkusDate: 2002-11-07 22:57:09
Subject: Re: Cannot assign ROWTYPE, RECORD variables in PL/pgSQL

pgsql-sql by date

Next:From: Prime HoDate: 2002-11-08 05:17:52
Subject: get source table.field name from view's field
Previous:From: Nekta KatzDate: 2002-11-08 00:24:12
Subject: Re: cast lo to oid

Privacy Policy | About PostgreSQL
Copyright © 1996-2017 The PostgreSQL Global Development Group