[OSM-dev] [OSM-talk] When mental models go wrong: OpenStreetMap

Dave Stubbs osm.list at randomjunk.co.uk
Fri Jul 4 09:14:06 BST 2008


On Fri, Jul 4, 2008 at 7:20 AM, m*sh <hy at sha-mash.de> wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Lauri Hahne wrote:
>> [...]
>>
>> 1. Ways
>>
>> - There are one kind of ways in the database. Ways can have attributes
>> (tags) but aren't required.
>> - The tagging scheme suggest there are many kinds of ways (highway,
>> cycleway, aeroway...)
>> - Potlatch and JOSM's preset (and mappaint?) system allow you to pick
>> a type for way.
>> - We have questions such as "how can I turn this way into a bridge" or
>> "how can I create a "bridge-way" at #osm once in a while.
>>
> [...]
>
> Ways are IMHO somewhat mis-designed; at that point I agree with you.
> i must admit that I have not too much of a deep insight into some parts
> of the data model - due to being involved since just 4 weeks now.
>
> Moreover, IMHO ways are completely obsolete. Every kind of attribute you
> need can be represented within a node. And it can be represented
> _easier_ by just attributing the nodes.
> Think of the third world, where roads are often not covered with tarmac
> - - the surface changes every now and then. With OSM you have to split it
> up in several ways, each attributed accordingly.
>
> All you would need is an item or *tag* k='connect' v='<node_id>'
> So you could connect nodes to a way, and you could easily turn two nodes
> within a road into a bridge if it has not been set appropriately before.
>
> Any renderer has to use the nodes anyway to make up a "way" so it has to
> look at the nodes - WHY look at and interpret the ways additionally?
>
> Admittedly any software would need a fix on that. But the major part
> (dropping the ways and setting the connections) should be a matter of a
> simple script.
>
> Just my 2 cents on your thoughts


Relational integrity, tag ambivalent database implementation, and a
vague chance of keeping history sane, plus just a little abstraction
to stop us all going completely insane?

But kudos to you, most people try to normalise the data model onto
relations, but normalising to nodes is thinking outside the box. Of
course it's basically a tag version of segments which were ditched due
to the extra processing requirements, and the annoying unordered ways
thing. And also I think you have a problem where a node is shared by
multiple "ways".. ie: where do I put my bridge tags?

Anyway... redesigning the raw data model is more than likely a
complete waste of everyone's time.

Dave




More information about the dev mailing list