On 08/09/2017 13:38, Juergen Schoenwaelder wrote:
On Fri, Sep 08, 2017 at 01:19:03PM +0100, Robert Wilton wrote:
In the same vane, I think that you could regard RFC 6087 and 6087bis as one
long list of CLRs ...
No. There are guidelines that have a clear technical reason. An example:
The 'preceding-sibling' and 'following-sibling' axes SHOULD NOT used.
A server is only required to maintain the relative XML document order
of all instances of a particular user-ordered list or leaf-list.
Yes, but even this rule has problems. Does this mean an implementation
does not need to support "preceding-sibling and following-sibling"?
Given this is only a "SHOULD NOT", it means that there might be some
modules may still use them. Likewise for the rest of the XPath "SHOULD
NOT" rules. Will YANG implementations fragment on which subsets of
Xpath they regard as sane and choose to implement?
[As an aside: I actually think that it would be better to restrict the
usage of Xpath to a much smaller subset that makes sense, but that
should be done in 7950bis rather than 6087bis.]
Besides, 6087bis has many softer rules as well, a few examples give
below (I'm not even sure why the last one is a rule). There don't
obviously appear to be any technical reasons for any of these, but I
don't object to any of these since they are provide sensible guidance,
or try to encourage consistent modelling conventions in IETF YANG models.
Ex1:
Identifiers SHOULD follow a consistent naming pattern throughout the
module. Only lower-case letters, numbers, and dashes SHOULD be used
in identifier names.
Ex2:
Definitions for future use SHOULD NOT be specified in a module.
Ex3:
The signed numeric data types (i.e., 'int8', 'int16', 'int32', and
'int64') SHOULD NOT be used unless negative values are allowed for
the desired semantics.
This is very different from guidelines how things should be named that
do not have a real technical reason. In SMIv2 land, we had this weird
rule that names of counters should end with a plural 's' and tools
started to implement this and to complain if there was no plural s.
(Actually, tools checked whether the last character is an s, which of
course does not mean there is a plural form.) And of course, there are
irregular nouns in English wrt. plural forms.
As per ex1 above, perhaps YANG tools will start to assume that
identifiers cannot contain uppercase characters.
It might be better if a lot of the guidance in 6087bis is changed to
avoid using RFC 2119 language precisely so that it can't be subsequently
interpreted as a formal rule.
But I still see guidance on how to consistently name counters and list
elements is good way of helping achieve consistency, and make the models
read better. This doesn't mean that tools should interpret these as rules.
I do not want rules that say '-state' should not be used. There may
be valid reasons to use '-state'.
Yes there might, but most likely when someone uses "-state" in the name
of a container they will be doing the wrong thing, and it may cause them
problems down the line. Warning them of the potential problems now so
that they make an informed decision seems generally helpful to
humanity. This does not mean that it needs to be a rule, or is even
allowed to be interpreted as such.
Thanks,
Rob
/js
_______________________________________________
netmod mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/netmod
