On Wed, Nov 30, 2022 at 4:08 PM Robert Sparks <rjsparks@xxxxxxxxxxx> wrote:
On 11/30/22 2:39 PM, Brian Campbell wrote:
Thanks for that pointer. If that's the extent, then I really think the group should walk this back just a little and answer why restricting these names to (a subset of) ascii is an unacceptable thing to do. The conversation there reinforces my guess that these aren't meant for display to users, so why take on the additional complexity? Make it easy for implementors to get it right with much less effort.Thank you for the review Robert. And apologies for the very delayed response. I think we had a bit of a volunteer's dilemma amongst the editors, which was exacerbated by some timing issues including vacation and subpar communication amongst us. We'll get all the nits/editorial comments addressed. Also the minor issues. Actually, changes for those are already in a branch of the document source repository https://github.com/oauthstuff/draft-oauth-rar/tree/genart-review and should be in the -17 revision. Some discussion on the majors is inline below.
On Thu, Nov 17, 2022 at 3:45 PM Robert Sparks <rjsparks@xxxxxxxxxxx> wrote:
<snip>
I have two major issues that I think need discussion:
Major Issue 1) The document seems to be specifying a new way of
comparing json names, claiming it is what RFC8259 requires, but I
disagree that RFC8259 says what this document is claiming. If I'm
correct, the document is trying to rely on the text in section 7 of
RFC8259 to support the idea that implementation MUST NOT alter the json
names (such as applying Unicode normalization) before comparison and
that the comparison MUST be performed octet-by-octet. Rather, section 7
says something more like "you better escape and unescape stuff correctly
if you’re going to do unicode codepoint by codepoint comparison" which
is a completely different statement.
If I'm right, and this is a new comparison rule that goes beyond what
JSON itself defines, I think the group should seek extra guidance from
Unicode experts. If I'm wrong and this behavior is defined somewhere
else, please provide a better pointer to the definition.
In many environments, its unusual for an implementation relying on a
stack below it to have any say at all on whether normalization is going
to be applied to the unicode before the application gets to look. Rather
than trying to work around the problem you've identified with
normalization by specifying the comparison algorithm, consider just
making stronger statements about the strings used in the json names the
document defines. Why _can't_ you restrict the authorization_details
values to ascii? If it's because you want to present the string to a
user, consider putting a presentation string elsewhere in the json that
is not used for comparison.
To the best of my understanding, it's not trying to specify a new or different way of comparing JSON names or values. I think it's only trying to say that the application must not do any *additional* normalization of the string values that it gets from the JSON stack or any other extra processing for the sake of comparison. I think anyway.
Honestly, I didn't really (and still don't) understand the concerns that some of the WG had that led to the text in question. So I didn't pay close attention to it while thinking to myself there can't be harm in saying to do a byte-by-byte comparison with no additional processing. But here we are...
Does that halfhearted explanation alleviate your concerns at all? Or, with that explanation in mind, are there specific changes to the text (in sec 12 and sec 2, I think) that would alleviate your concerns? Or do we need to consider just deleting those parts?
I did track down this issue about it https://github.com/oauthstuff/draft-oauth-rar/issues/28 for maybe added context.
Yes, that guess is correct that type value is not meant for display to users (the issue is about type value). I can't confidently say the same about names and values that any particular type will define. I don't think it matters though. It's not that restricting is unacceptable but that it's not necessary. Just using the values that come from the JSON layer is enough. And I'd contend there's not really additional
complexity. It just appears like additional complexity because of the unnecessary and maybe less than idea wording in the draft. Wording that I'd be more than happy to try and fix up or just remove.
This works for me. Consider adding something like "This mechanic is illustrative only and is not intended to suggest a preference for designing the specifics of any authorization details type this way."
Major Issue 2) The suggested pattern demonstrated starting in figure 16
(using [] to mean "let the user choose") seems underspecified. If the
point is that different APIs may invent different mechanics _like_ this,
and that this is only an example. Make it much clearer that this is an
example. If this is a pattern you expect all APIs to follow, then more
description is warranted. Is it intended that a user could add and
remove things arbitrarily to such lists? For instance is it intended
that this support an interaction where the client is asking for
permission to operate on account A, and the user can say "no, but you
can operate on account B"?
It is really intended to be saying that different APIs may invent different mechanics _like_ this, if needed. And the []'s are just one way that an API might define some of how to do it.
I've tried to make this more clear with these edits: https://github.com/oauthstuff/draft-oauth-rar/commit/ee70e000557a69afe133356847c5083882686811
Sure, I'll add that.
CONFIDENTIALITY NOTICE: This email may contain confidential and privileged material for the sole use of the intended recipient(s). Any review, use, distribution or disclosure by others is strictly prohibited. If you have received this communication in error, please notify the sender immediately by e-mail and delete the message and any file attachments from your computer. Thank you.
-- last-call mailing list last-call@xxxxxxxx https://www.ietf.org/mailman/listinfo/last-call
