Re: References and Modularity

On Tuesday, June 4, 2013 at 2:09 AM, Karl Dubost wrote:

>  
> Dated versions, md5 hashtags, any kind of schemes that identify a specific version in time is irrelevant in that discussions. Versioning systems are exactly the same thing than a dated space with smaller increments.  
>  
> The interesting questions are along these axis:
>  
> 1. How does a user (implementer) find the last version?
There is only one URL to get at a spec (or a few short hands, but all point to the same thing - like whatwg.org/c does).   
> 2. How does a user (historian, lawyer) find the nth version?

For lawyers, you point them to a particular commit in the version control system, as we do with CGs. For example:
https://dvcs.w3.org/hg/html-proposals/raw-file/9443de7ff65f/responsive-images/responsive-images.html  

(where "9443de7ff65f" is the relevant commit)

For historians: they access the version control system history (CVS, GIT, HG, whatever) + bug repo + mailing list. I can't think of better archeological tools.  
> 3. How does the document informs users that a nth version is not anymore the one to follow for his/her specific usage?

There is only one version - it's marked as "Living Document". That sets the expectation that it is maintained. User's don't go to Google and seriously expect to access the same search box that was there 8 years ago; it's the same with specs. I don't go to the HTML spec and expect it to be the same as it was yesterday … or else Hixie and the rest of the Editorial team on the W3C side are slacking off! :)   
> 4. What are the expectations for this version with regards to certain constraints and users.  

> - implementers
> - lawyers  
> - users
> - etc.
>  
> The issue with dated versions at W3C in the current context is not the fact of using/pointing to a dated version, but what *some* dated versions mean in the process.
Right. If you look at HTML5 and some of the newer CSS specs, you see the *BIG RED WARNING* to get around this problem with the process.  

So, as HTML5 shows, a process that relies on dated versions is a broken process. This problem seems to stem from the W3C process being model on a pre-digital-publishing standardization model (i.e., from a kind of "nuts-and-bolts" standardization process, where standardized products remain static and don't get changed after production). That doesn't apply to us.  
> In the things we do not do very well at W3C (and any versioning systems) is linking to n+1 (forward linking).
>  

IIUC, the problem can be circumvented by having a single point of reference, a little cache control, and some kind of server-sent event. This exists today: the WHATWG HTML spec notifies the reader if the version they are viewing gets updated (this might even happen in real time).

Received on Tuesday, 4 June 2013 10:22:03 UTC