Martin Bjorklund je 17.3.2016 ob 21:32 napisal:
"Larsson, Gustav" <[email protected]> wrote:
(Originally posted to the netconf mailing list in error. My
apologies.)
Hello,
When a node defaults to config true and a deviation changes it to
config false, should "deviate add" or "deviate replace" be used? Some
YANG implementations allow either, some allow only "deviate
replace". Perhaps some allow only "deviate add" but we have not yet
seen that.
Consider these definitions:
container x {
leaf y {
type string;
}
}
deviation /x/y {
deviate replace { <-- Should this be add or replace?
config false;
}
}
Section 7.18.3.2 of RFC 6020 says:
The argument "add" adds properties to the target node. The
properties to add are identified by substatements to the "deviate"
statement. If a property can only appear once, the property MUST
NOT
exist in the target node.
All nodes in the data tree has a "config" value; if it isn't
explicitly specified it is inherited from the parent, and if the
top-level node doesn't have an explicit statement it defaults to
"true". So in order to change this property, it should be deviated
with 'replace'.
The argument "replace" replaces properties of the target node. The
properties to replace are identified by substatements to the
"deviate" statement. The properties to replace MUST exist in the
target node.
/x/y is config true by default even though config is not explictly
given. Section 7.19.1 of RFC 6020 is not clear about whether config is
considered to exist when defaulted.
Don't you think it is clear how the config value is inherited?
If "config" is not specified, the default is the same as the parent
schema node's "config" value. If the parent node is a "case" node,
the value is the same as the "case" node's parent "choice" node.
If the top node does not specify a "config" statement, the default is
"true".
This text doesn't use the words "config property", this could probably
be clarified.
What needs to be clarified is the definition of "property". I always
read this as "substatement": "units", "default", "require-instance",
"when", "if-feature", "ordered-by" are all being referred to as
"properties" and those are not inherited. On the other hand, "mandatory"
is never referred to as a "property".
Also, "refine" statement text, which is practically analogous to
"deviation", explicitly uses "statement" for refinements (config included).
Our implementation requires "add", if no "config" substatement is
present in the target node (as did early versions of pyang, if I'm not
mistaken).
Jernej
/martin
If config is considered to exist, then "deviate replace" is
correct. Otherwise, "deviate add" is correct. Technically, RFC 6020
allows only one or the other but not both in this case, but perhaps a
robust implementation should accept both.
YANG file authors need to pick one or the other, so clear guidance
would be helpful.
Thanks,
Gustav
_______________________________________________
netmod mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/netmod
_______________________________________________
netmod mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/netmod
_______________________________________________
netmod mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/netmod
