On Wed, 2018-10-10 at 15:23 +0000, Michael Rehder wrote:
> Container "foo" would be mandatory if not for the "when" child element.
> With the "when" child element, the logic becomes "inverted" and the constraint
> is a negative one of "disallowed under certain condition".
>
> The UC is for enforcement in REST API payloads.
> For a practical example:
>
> leaf AssignmentMechanism {
> type enumeration {
> enum "DHCP";
> enum "Static";
> }
> mandatory true;
> description "The address assignment mechanism.";
> }
> list IPAddresses {
> when "../AssignmentMechanism = 'Static'";
> key Address;
> min-elements 1;
>
> leaf Address {
> type capit:IPv4Address;
> description "An ipv4 address.";
> }
> }
>
> There is no way in the IPAddresses list to enforce that there is at least one
> IP Address when the assignment method is "Static".
Why do you think there is no way? For example, according to sec. 10.28 of RFC
6110, then "min-elements 1" statement is mapped to the RELAX NG pattern
<oneOrMore>
which enforces the presence of at least one entry of IPAddresses.
>From the DSDL point of view, the problem is rather the opposite: at least one
entry will be required during RELAX NG validation even if AssignmentMechanism is
"DHCP".
Lada
> One could put a "must" on "AssignmentMechanism" to ensure at least one element
> of the IPAddresses list when "Static", but I don't see this as a good schema
> design, to have the controlling attribute check controlled attributes.
>
> I appreciate that this semantic can't be changed in YANG at this point.
> Could the "when" statement have a modifying child element to state that the
> mandatory status of the element is to be enforced?
> Like
> container foo {
> when "condition" {
> enforce-mandatory-status;
> }
>
> There is already back-end for existential checks for mandatory choice so this
> seems reasonably consistent to me.
> I appreciate there are existing issues for "when" but I don't see why this
> would make things any worse.
> In fact by promoting a better dependency "direction" between schema
> elements, think it could simplify things (so I naively think :) ).
>
> Thanks
> Mike
> > -----Original Message-----
> > From: Ladislav Lhotka [mailto:[email protected]]
> > Sent: Wednesday, October 10, 2018 10:28 AM
> > To: Michael Rehder <[email protected]>; [email protected]
> > Subject: Re: [netmod] WHEN statement within mandatory objects doesn't
> > ensure presence of the mandatory object
> >
> > Michael Rehder <[email protected]> writes:
> >
> > > I have a question about “when” and mandatory objects.
> > >
> > > It seems to me that the implemented semantics of “when” are really
> > “optional when”, in that the enclosing object can be absent even though it
> > is
> > mandatory and the “when” clause holds true.
> > > The RFC could be clearer about this.
> > >
> > > Example
> > >
> > > leaf color {
> > > enumeration {
> > > enum “blue”;
> > > enum “black”;
> > > }
> > > mandatory true;
> > > }
> > > container foo {
> > > when ../color = ‘blue’;
> > > etc.
> > > }
> > >
> > > “foo” is optional due to the presence of the “when” statement even
> > > though the object is mandatory (same is true for mandatory leaf,
> > > min-elements=1 list etc.).
> >
> > Maybe you intended to have, e.g., a "mandatory true" leaf inside "container
> > foo"?
> >
> > > This is considered valid XML for the above
> > > <color>blue</color>
> >
> > Yes, it is, under current YANG rules, no matter what "etc." stands for. Note
> > that
> > evaluation of the XPath expression in this case (with "foo" missing)
> > requires the
> > peculiar procedure of sec. 7.21.5 in RFC 7950.
> >
> > > In my view this makes conditionally variant schemas “loose” in their
> > > enforcement (some scenarios can use choice but it doesn’t cover
> > > everything).
> > >
> > > I think that mandatory should be respected for the enclosing objects
> > > of a “when” statement. That is, a mandatory object must be present
> > > when its “when” clause holds true and a Schematron statement should
> > > enforce that.
> >
> > In fact, this is one case where the DSDL mapping (RFC 6110) deviates from
> > YANG 1.0. Nodes that mandatory aren't enclosed in the RELAX NG <optional>
> > pattern, and are then required no matter what any "when"
> > statements say (because RELAX NG validation comes before Schematron).
> >
> > > What is the rationale behind the current YANG rules behavior, that the
> > > “when” Schematron mapping doesn’t check for presence of the enclosing
> > > mandatory object?
> >
> > FWIW, I have been repeatedly protesting against this behaviour but without
> > much luck. See for example
> >
> > https://www.ietf.org/mail-archive/web/netmod/current/msg14012.html
> >
> > As a result, "when" is the trickiest feature in YANG by far.
> >
> > Lada
> >
> > > thanks
> > > Mike Rehder
> >
> > --
> > Ladislav Lhotka
> > Head, CZ.NIC Labs
> > PGP Key ID: 0xB8F92B08A9F76C67
> “Amdocs’ email platform is based on a third-party, worldwide, cloud-based
> system. Any emails sent to Amdocs will be processed and stored using such
> system and are accessible by third party providers of such system on a limited
> basis. Your sending of emails to Amdocs evidences your consent to the use of
> such system and such processing, storing and access”.
--
Ladislav Lhotka
Head, CZ.NIC Labs
PGP Key ID: 0xB8F92B08A9F76C67
_______________________________________________
netmod mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/netmod
