Re: TMQL: Direct (super|sub)types?

On Sat, Mar 05, 2011 at 12:44:37PM +0100, Robert Cerny wrote:

> > Pragmas have been _part_ of the language. For exactly the reasons you
> > pointed out.
> 

> Ok. After reading the relevant sections in the 'old' TMQL proposal,
> i understand, that the pragma is only valid for one query and that
> there is no distinction between direct and transitive.

Yup.

> If i would
> wildly guess, i would say, because you think, that it *is* the same
> axis, so why give it a different name, thus also the necessity to
> specify the direction. This is a very consequent and clear
> approach.

Probably :-)

> Yet, when i wrote, down some queries a while ago this way, i was
> always unsure whether to use '<<' or '>>'. It was always a matter of
> trial and error. I had the same problem when reading queries.

That is a valid concern, albeit not in all cases:

    $t >> supertypes

is rather self-evident. OTOH,

    $a << types

will make you think for a second, or two. For _exactly_ that reason, 
a beautiful editor proposed a shortcut notation for these occasions:

    $a >> instances # which is the same as $a << types

I wonder what the alternative would be? Name not only the axes
differenly, but also every direction? With role types and variants we
would end up with ~30?

> wonder how this discrepancy comes about: on the abstract level, it
> makes so much sense, but on when applied, it is so difficult. But is
> my guess correct, that certain (if not all) 'old' TMQL path
> expressions can be understood no matter whether you start at the
> left or the right?

Yes, the linear way to express a filtering constraint over a
collection allows to be read l->r and l<-r.

\rho