[OSM-talk] conclusions from my postgres / postgis experiments

David Earl david at frankieandshadow.com
Thu Nov 23 10:40:10 GMT 2006

> I'd agree with your overall comments here and I believe it would
> be counter
> productive to have editors strictly stick to linear ways. However
> it should
> be possible to break a complex way down automatically into its constituent
> linear parts and then place them all in a reference container
> (superways has
> be coined). The end users don't need to see this but it would
> make rendering
> and further usage of the data more straightforward.

Indeed. One way of doing this from the existing data is by looking for
connected ways with the same name. But should the database actually
represent this explicitly or should the tools that use it interpret
connected commonalities (superways, features with the same name that share a
node) in their own way? (Searching by name is one such tool, which can
clearly kill connected commonalitites easily; a renderer is another such

Superways have problems. For example, the whole A1134 (our outer ring road)
in Cambridge should be some kind of superway. But it changes name as it goes
along, so each of those pieces should be superways, suggesting a hierarchy
of superways. But those in turn have branches off them which have the same
name but aren't part of the A1134 (they are a residential spur which it
wasn't felt necessary to have its own name but is not part of the A1134 -
doesn't have 'primary' status) - so superways are actually a set of
overlapping sets, and how those groupings are managed depends on the use to
which the data is being put, whch is why I think it might be better to leave
the interpretation to the application. Though I suppose if it is explicit,
they can be treated as hints.

But in any case, the concept has to be multilevel and overlapping.

> the tree-like things are easy to handle, and can be transformed into
> multilinestring.
> as soon as you start having loops in there though, the problem becomes
> non-trivial

Perhaps, but that's the real world for you. But it depends what you are
trying to do with the data. The key aspect for me is how you would render a
name sensibly in the face of disconnected or disordered segements, branches
and lopps. I can think of a number of ways I could do that which aren't all
that complicated.

It's 3-dimensional too when you start taking into account flyovers, bridges,
aqueducts and the like.


More information about the talk mailing list