> From: Edward C. Zimmermann
> but does
>   mm mean month mm or day mm or century?
> In we have YY for "century" where "19" is given as an example.

'xx' as a standalone expression means century (and note that we have
profiled this out).  However, in an interval, I believe that given the lack
of clarity in 8601 we have the latitude to interpret unclear areas (where no
external guidance is available) as we see fit, and I propose to interpret
8601 as not allowing centuries in intervals.

  So 'mm' means month or day depending on the context. Thus:

1984-04/06   April through June of 1984. 

1984-04-04/06   the 4th through 6th of April 1984.

> Why should we assume, as they have here, that the year implied is the
> year of the start of the interval?
> If "1985-04-12/06-25" means 12 April 1985 to 25 June 1985 what does
> "1985-04-12/02-25" mean?  12 April 1985 to 25 Feb 1986?

Good, question, and I would say yes, that's what it means. 

> And "1985-04-26/25"?  26 April 1985 to 25 May 1985?

Yes, I suppose so.  But then what does "1985-04-26/26"  mean?  26 April 1985
to 26 May 1985, or 26 April 1985 to 26 May 1985?

> That is how my parser currently behaves. An interval specified as from
> 12 Oct 1985 to Jan is interpreted as 12 Oct 1986 - Jan 2011. 

I don't believe that's a meaningful interval. The start and end need to have
the same "precision", implied or not,  don't they?

I read
> each point, start and end, as dates and don't assume the end or start
> depends upon the information content of the other...

Well, to some extend we have latitude to interpret as we like but there is a
limit to creative interpretation. If we are going to use the example in 8601
as evidence that they intended to allow intervals without the time
component, then we cannot just ignore the fact that that same example allows
omitted/implied values. 

> 18850203/20 I would read as 3 Feb 1885 through the "century" defined by
> 20 rather than, as in the example, the 20th of Feb 1885.

I would prefer to assume that century is not intended to be used in an

>  its less than satisfying to
> guess the intent... If that is what 8601 wants.. I'd say we toss it
> out.. and adopt the--- and least what I have considered--- consistent
>   start/end
> where start and end are read as 8601 dates (resp. our extended forms)
> and consider intervals where end>start as.. bad intervals...
We can do this. We could leave intervals completely out of level zero and
put all "normal" intervals in level 1. 

I'd like to hear opinions on the direction we should take.

And the more I think about it the more I am convinced that we should take
this approach, do not allow inferred date components, so that we don't have
to come up with a load of complex rules.