On Mon, Mar 6, 2017 at 8:20 PM, Brian E Carpenter <brian.e.carpenter@xxxxxxxxx> wrote:
On 07/03/2017 17:06, Eric Rescorla wrote:
> On Wed, Mar 1, 2017 at 11:24 AM, Brian E Carpenter <
> brian.e.carpenter@xxxxxxxxx> wrote:
>>
>>> I'm not arguing against updating data tracker more often - just saying
>> this
>>> 'editor's draft' convention can work very well between official revisions
>>> no matter the cadence a WG chooses.
>>
>> The details of that discussion probably belong on ietf-and-github@xxxxxxxx
>> ,
>> but I must point out that this way of working *excludes* from the
>> discussion WG participants who don't grok github. Substantial issues
>> need to be discussed on the mailing list and substantial (non-typo)
>> revisions need to be posted as I-Ds.
>>
>
> Well, it's hard to know what to make of this without knowing what you
> mean by "substantial" but an active draft takes literally hundreds of PRs
> in its lifetime with perhaps half of those being non-typos. We could of
> course gateway every PR merge to an IETF draft push. Is that what you're
> looking for?
No. In fact (countering Michael's point slightly) I get quite annoyed
by draft versions that turn out only to fix few typos or grammatical errors;
those can wait. As for what constitutes "substantial", that's very subjective.
Anything that causes an on-the-wire protocol change would certainly be
substantial. Clarifying ambiguous text might be substantial. But YMMV.
But this would still lead to a colossal number of drafts. It's easy to see this by
looking at recent versions of TLS, where I have handily noted all on-the-wire
changes in the changelog (or at least tried to). Here are the numbers from
recent drafts:
Draft Wire changes
-19 (unpublished) 5
-18 1
-17 9
-16 6
-15 5
-14 5
As you can see, we have an average of 5 wire-level changes per revision,
so given that we are at draft-19, we would be at around draft-100 by now.
Probably more, because we were more aggressive earlier, and I just
started noting these because people were fielding implementations and
I wanted to make it easier to update. I find it hard to believe that people
would in fact be happier if TLS 1.3 were currently at -100 and to the
best of my knowledge, next to no WG has every produced drafts
at anything like that frequency.
Note that I'm not saying that TLS is especially active (though I suspect
it is more than average). Rather, people (both with and without Github)
tend to batch up a bunch of changes and submit them in a single draft.
It's just that because Github makes the revision history so public and
explicit that people are aware of these as discrete events where you
could publish as individual drafts, rather than as a pile of things that
go into a single draft.
-Ekr