TAP-1.1 PR is now available

[Markus Demleitner]

Hi again,

On Mon, Sep 11, 2017 at 11:09:27PM +0100, Mark Taylor wrote:
> Markus's (post PR-20170830) rev 4228 edit for TAP.tex includes the
> comment:
> 
>    "Note that while TAP 1.1 does not require the use of any
>     particular minor version of the standards it depends on..."
> 
> I must admit that I don't know whether that's supposed to be
> the case or not, but if so it would suggest that VOTable 1.3

Well, I believe for interoperability we should avoid mandating
specific versions of dependencies, treating all minor versions of a
standard as equivalent.  I'd make a point that that's essentially
implied by the semantic versioning we're claiming to have (somewhere
in DocStd).

Yes, that requires a certain amount of care, as for instance with
VOTable 1.3, where we said clients need to explicitly ask for BINARY2
if they want it (which certainly hasn't bolstered its adoption).  But
I'd argue it's well worth it -- without it, we'll quickly be in a
tangle of mutual requirements that will make evolution very difficult
and slow indeed.

> is not be mandated.  I don't have strong opinions either way,
> but I would welcome explicit clarification in the document text
> about whether TAP does make requirements about minor versions of the
> standards it mentions (DALI, UWS, VOTable, VOSI, VODataService,
> ADQL, more?), either globally or per-standard.

I don't see a major case for diverging from the general rule of
minor-version-blindness in any of TAP's dependencies.

*If* we were to pragmatically decide to diverge, my first choice
would be UWS 1.1.  The "long" polls are a big help.  But then if
clients don't immediately re-poll but wait a couple of seconds, my
guess would be that blindly trying long polls on UWS 1.0 shouldn't
hurt either.  So, even there I'd say there's no strong reason.

In the VODataService case I'm confidently planning on re-allowing
tables outside of schemas in 1.2.  This feels a bit as if it could
throw clients off.  But that would be a VODataService problem and
shouldn't be fixed in TAP by restricting to 1.1.

        -- Markus