[OSM-talk] Advanced relationships

[Ian Sergeant]

Frederik Ramm <frederik at remote.org> wrote on 14/06/2007 04:27:03 PM:

> > The only remaining question is how to represent the segment pair.  My
> > suggestion is to use a way, since that is the only current way to link
two
> > segments.  A new way tag would be needed.
>
> Come on, it isn't as if we need to express everything there is in ways
> or segments or nodes!

A mechanism to link segments is all that is required.  Currently a "way" is
a set of linked segments which provides a rendered path.  It isn't too much
of a corruption to think of a "way" as just one specialised example of
segment links.

> ... little patches and hacks are not the way forward. This will end
> in a data set where half of the ways and tags are not really ways or
> tags but just some kind of helper constructs to circumvent the weakness
> of the data model.

It really depends what you think of as a "way".  We need to consider the
data structure itself and the name it is given separately.  It may not be
sensible to change the data model to add another segment linking data
structure, and call it something else when it has the same data properties
as the existing "way" data structure.  Especially when all the restriction
tags will equally apply.  All that may be required is making the current
use of a "way" a specialised use of a more general (existing) data
structure.

> We need proper relationships and not a clever abuse of existing
mechanisms.

I would also support going down this path if sufficient momentum is there
to change the data model.  The most elegant solution achievable should be
the target.  At the low end of the elegance spectrum is actually corrupting
the physical representation in order to implement routing restrictions.
This is happening now. At the high end is new data model designed from
scratch, and this seems to me (given the recurrence of this discussion) to
be a long way off.

Ian.