Re: Multiple "similar" controls per op

[mcs@netscape.com (Mark C Smith)]

"Kurt D. Zeilenga" wrote:
> 
> At 08:26 PM 3/8/01 -0700, Jim Sermersheim wrote:
> > RFC 2251 allows one to specify mutliple controls on a single
> > operation. It doesn't disallow two or more controls of the same
> > type to exist on the operation. This means we either need to:
> >
> >a) Add text to the "recommended control specification guide" (or
> > whatever it's called), that tells authors of control spec's to call
> > out whether this is allowed with that particular control. Or,
> >b) ammend 2251 to disallow this.
> >
> >I lean towards a), just to leave the door open. How do others feel.
> > If b), we'll take it to the LDAPBIS list.
> 
> I concur.  I believe it unwise to place unnecessary limitations
> upon the kinds of extensions which might be defined.  Hence, a)
> seems like the most appropriate option.

I also agree.  While I see no strong reason for using two controls
instead of, for example, a sequence within the value of a single
control, I also see no reason to prevent this if someone wants to design
an extension that way.  It would be interesting to see how various
implementations behave if they receive two controls with the same OID
today.  Possible behaviors:

1) Ignore all but the first instance of the control.

2) Ignore all but the last.

3) Generate and return an error (which one?)

4) Something else.

So we may want to provide some advice to server implementors somewhere
about what to do if a client sends two instances of a control (the best
behavior may depend on the control though, e.g., two manageDSAIT
controls vs. two server side sorting request controls). 

-Mark