JSON Web Signature (JWS) Unencoded Payload Option
draft-ietf-jose-jws-signing-input-options-09

-7, last paragraph: 

" Thus, method 1 -
   requiring support for this extension - is the preferred approach and
   the only means for this extension to be practically useful to
   applications."

One might wonder why method 2 and 3 are included. I assume it is to allow existing apps to migrate to method 1 over time? If so, some guidance on app migration might be useful.

Editorial:

-6, last paragraph:
It’s confusing to see "(JWT) [JWT]" . I suggest either removing (JWT), or changing the anchor for the citation to use [RFC7519]

As mentioned by Stefan in his OPS DIR review:
There are coexistence issues between implementations which understand
the notion of the "b64" parameter (i.e. implementing this RFC) and those
who do not (i.e. all existing JWS implementations).
The issues are discussed in Security Considerations (second para up
until the end). The issues with it are:

* the conclusions are not as satisfactory as they could be.
Interoperability is either

 - left as an out-of-band and undescribed mechanism ("make sure that
only supporting implementations talk to each other")
 - by explicitly marking support for b64 as critical (IMO the only good
solution)
 - modifying the payload so that it contains unparseable characters
(which may or may not be possible for the sender depending on the
payload characteristics)

* this is placed in Security Considerations while it has actual
operational impact on every transmitted message: in essence it states:
"Sometimes, sender and recipient may misunderstand each other without
noticing". Example given in the draft where the actual message is "NDA1"
while the recipient thinks it was told "405", without a way to notice.
Even if the misunderstanding is not related to security, it can/will
have significant implications for the application.

I believe this can not be left as-is. Luckily, there seems to be an easy
way out:

"Whenever the 'b64' header exists and is set to false, the 'crit' header
MUST also be present and contain 'b64'."

This, maybe in conjunction with

"When the content is intended to be base64 encoded, the 'b64' header
SHOULD NOT be present."

This would make sure that implementations who know nothing of b64 are
left alone (there is no unknown extension, there is no criticality in
any such extension, and the sender did not intend to make use of the
feature => all good). While at the same time for unencoded payloads
making deterministically clear that unencoded transmission is happening,
and is required to be understood.

This would at the same time make a complex piece of Sec Con go away.

I am a no-objection on this, but I do agree with Stephen's Discuss
position, raised in the Gen-ART review by Robert.

stefan winter performed the opsdir review