Re: [regext] AD Review: draft-ietf-regext-org-ext-07

[Adam Roach <[e-mail]>]

Responses inline.

On 7/30/18 1:21 AM, Linlin Zhou wrote:
> Dear Adam,
> Thanks for your review. I have my feedbacks started with [Linlin]. 
> I'll update the draft based on your comments.
>
> Regards,
> Linlin
> ------------------------------------------------------------------------
> [e-mail]
>
>     *From:* Adam Roach <mailto:[e-mail]>
>     *Date:* 2018-07-28 07:04
>     *To:* draft-ietf-regext-org-ext
>     <mailto:[e-mail]>; [e-mail]
>     <mailto:[e-mail]>
>     *Subject:* [regext] AD Review: draft-ietf-regext-org-ext-07
>     This is my AD review for draft-ietf-regext-org-ext-07.  I have a
>     handful of
>     comments below that I'd like to see addressed prior to asking the
>     IESG to
>     consider the document. Please treat them as you would any other
>     last-call
>     comments.
>     There are also two blocking comments that need to be resolved
>     prior to
>     IETF last
>     call.
>     Thanks to everyone who worked on this document.
>     ---------------------------------------------------------------------------
>     This is a blocking comment.
>     This document raises the same question as draft-ietf-regext-org-08
>     does
>     regarding the allowable placement of XML namespace declarations
>     within the
>     document; see, e.g., the following text:
>     >  In addition to the EPP command elements
>     >  described in the EPP object extensions, the command MUST contain an
>     >  <extension> element, and the <extension> element MUST contain a
>     child
>     >  <orgext:create> element that identifies the extension namespace if
>     >  the client wants to associate data defined in this extension to the
>     >  object.
>     I presume the same answer will apply to this document as does to
>     draft-ietf-regext-org-08.
>     Affected elements appear to also include <orgext:update> and
>     <orgext:infData>.
>     [Linlin] Please see my feedback in the reply of org draft. Thanks.
>


I assume we'll resolve this the same way in both documents.


>     ---------------------------------------------------------------------------
>     This is a blocking comment, as it impacts interoperability.
>     §4.2.5:
>     This section defines remove and change elements that use "role" as
>     a key. It
>     is unclear whether an attempt to remove or change an identifier
>     corresponding to
>     a role that is not present in the object results in an error, or is
>     treated as
>     success.
>     For example, if an "example.com" is currently in the system as a
>     reseller, but
>     is *not* in the system as a privacyproxy, would an update
>     containing the
>     following elements return a success response or an error response?
>        C:      <orgext:update
>        C: xmlns:orgext="urn:ietf:params:xml:ns:orgext-1.0">
>        C:        <orgext:rem>
>        C:          <orgext:id role="privacyproxy"/>
>        C:        </orgext:rem>
>        C:      </orgext:update>
>     If the answer is that an error is returned, then that error needs
>     to be
>     clearly
>     specified in this document.
>
>     [Linlin]I think an error should be returned.
>


Okay -- we need to say which error code to use, then.


>     The same question needs to be answered for <orgext:chg>.
>


Is the answer the same for <orgext:chg> as for <orgext:rem>?


>     Similarly, if <orgext:add> is issued for a role that already
>     exists in the
>     object, does this result in an error, or is the existing role
>     identifier
>     silently overridden?
>


This question also needs an answer.


>     If the answer to "is this an error" is "yes" for any or all of the
>     preceding questions: this document needs to clearly spell out what
>     happens when
>     an <orgext:...> element contains multiple <orgext:id> elements, and
>     *some* of
>     them cause an error while *some* of them do not.
>


This also still needs to be addressed. For example:

If "example.com" is currently in the system as a reseller, but is *not* 
in the system as a privacyproxy, what would the following command do?
    C:      <orgext:update
    C: xmlns:orgext="urn:ietf:params:xml:ns:orgext-1.0">
    C:        <orgext:rem>
    C:          <orgext:id role="reseller"/>
    C:          <orgext:id role="privacyproxy"/>
    C:        </orgext:rem>
    C:      </orgext:update>


This could do any of the following, and the document needs to be clear 
which one actually happens:

 1. The command succeeds, and the "reseller" ID is removed from
    "example.com"
 2. The command fails because "privacyproxy" doesn't exist as an ID on
    "example.com." No changes are made.
 3. The command partially succeeds: the "reseller" ID is removed from
    "example.com," but the response is an error message because
    "privacyproxy" could not be returned.

The semantics around #3 are very complicated, since you'll ultimately 
need to indicate which part of the command succeeded and which part 
failed, so you probably want to pick #1 or #2. Given your answer above 
that removing a non-existent orgext-ID from an object is a failure, I 
think #2 is the most consistent. But this needs to be clearly specified.



>
>     Finally, if the <orgext:add> and <orgext:chg> elements do not
>     result in
>     errors
>     in the cases described above, then this document should clearly
>     specify how
>     processing is different between those two elements, or clearly
>     specify that
>     handling of both elements is identical.
>     [Linlin] So is it ok to add some words like "An EPP error response
>     MUST be returned if an <update> command cannot be processed for
>     any reason." ?
>


That's really not enough. You need to be very clear about what "cannot 
be processed" means. And, since you have commands that perform more than 
one operation at the same time, you need to be very clear about handling 
when one of those operations would be okay, but the other one is not.

I don't want to tell you how to resolve each of these issues; but, based 
on the one answer you gave above (about removing a non-existent ID), the 
following clarifications would be consistent:

 1. An attempt to remove an ID that does not exist results in an error
    with a result code of UUUU
 2. An attempt to change an ID that does not exist results in an error
    with a result code of VVVV
 3. An attempt to add an ID that *does* already exist results in an
    error with a result code of WWWW
 4. An attempt to remove more than one ID where at least one of them
    does not exist does not change the object at all, and results in an
    error with a result code of XXXX
 5. An attempt to change multiple IDs where at least one of them does
    not exist does not change the object at all, and results in an error
    with a result code of YYYY
 6. An attempt to add multiple IDs when at least one of them already
    exists does not change the object at all, and results in an error
    with a result code of YYYY

You will need to say all six things. Also, for #4, #5, and #6, you'll 
need to think about whether there is any way for the client to know 
which ID caused the operation to fail

Note that the result codes above might be the same as each other or 
different from each other. I have no opinion on which is better, as I'm 
not familiar with the philosophy of how result codes are used in EPP.



>     ---------------------------------------------------------------------------
>


The resolution to the remaining issues all seem fine to me. Thanks.

/a