Re: New Version Notification for draft-pardue-http-identity-digest-01.txt

Hi Roy,

Thanks for reading and providing some feedback!

On Wed, Mar 8, 2023 at 5:39 PM Roy T. Fielding <fielding@gbiv.com> wrote:

> This misuses the term identity, which was a poor use of a word in the
> first place to refer to a transformation that doesn't change the content.
> In any case, "identity" refers to the coding, not the content.
>
> It doesn't help to use the same term for two different things in HTTP.
>
> If we want to send this field in a message with Content-Encoding applied,
> then it would be easier understood as Decoded-Digest or
> Digest-After-Decoding
> or Want-Unencoded-Digest (or maybe just Digested and WUD -- I hate
> long field names).
>

I won't argue here. I mainly picked identity as continuity from past
discussions (that also misused it). Totally happy to bikeshed and refine my
use of terms.


> Multiple digests should not be sent in the same message unless their
> digests are different. Transforming the message content to remove the
> outer coding would necessitate replacing the field values as well, as in
> replacing the Content-Digest field-value with the Identity-Digest's value
> and removing Identity-Digest. I know it's a cute example, but in my
> experience the readers tend to implement the examples in practice
> even when they are nonsensical and a waste of network bandwidth.
>

Can you elaborate some more?

I like the message transformation example.

We already allow the same value to be sent in Content-Digest and
Repr-Digest. Are you saying that Content-Digest and Identity-Digest have a
different relationship that needs other considerations?

Tweaking the examples to highlight the benefits and avoid implying certain
practice is a good suggestion I'll take.



>
> Also, you have not considered multiple content codings yet, which would
> (rarely) consist of a private encoding enclosed within a compression:
>
>    Content-Encoding: bzerp, gzip
>
> I would expect, in that case, for both encodings to be decoded before the
> identity-digest is applicable. YMMV.
>

I did consider that but sounds like I didn't articulate it in the doc. I
agree in general and will add some text. I don't think a private encoding
needs to be particularly called out though, I've seen some responses that
are double-gzipped for example.

Cheers
Lucas