Hi Ricardo & Brian,
On Mon, Oct 30, 2017 at 5:10 AM, Riccardo Bernardini <framefritti@xxxxxxxxx> wrote:
On Mon, Oct 30, 2017 at 4:48 AM, Brian E Carpenter <brian.e.carpenter@xxxxxxxxx> wrote:"
My main concern with the new proposal is other:
> 1. Is it stable, mature, and immutable (except for errata)?
Those are not well known characteristics of open source projects.
Even if we tie a reference down to a very specific version,
I'm unclear how that can truly act as a stable reference. It may
be stable in a mathematical sense, but it may also be seven years
behind deployed reality.Indeed, this was my first thought too. Although I like open source software (and I find it often better than the proprietary counterpart, although with maybe less bells and whistles), it is true that documentation in open source projects have a huge variability (I am currently trying to write a Linux module to be loaded at run-time and the biggest obstacle is the scarce documentation... Moreover, Linux kernel API can change from one release to the other, making that tutorial written five years ago totally useless today...)
Sadly, our RFCs can also be years ahead or behind of reality. We also have the ability to Update and Obsolete RFCs - with consequently impacted referring documents.
In our draft, one requirement is that the work be intended to be a public interface and another is that it is mature work.
Open source projects can vary from "I coded this up yesterday - go play" to extremely mature projects that run parts of the Internet. A conversation needs to recognize this range and the nuances.
Since the main feature of protocols should be interoperability, I think that a formal and unambiguous specification (and not just "as the GNU implementation of version 3.2.4 of Foo protocol behaves") is necessary, unless the cited open source protocol plays a very marginal role. Maybe a solution could be to take a "snapshot" of the protocol in an Informational RFC document to be written, possibly, in collaboration with the OS community. However, with this approach it seems to me that we are back to the downref process.
In general, the draft recommends using specifications. There can be cases - such as algorithms - where the code is really the best descriptor.
I'm happy to hear suggestions for improving the text - of course!
Regards,
Alia
Riccardo