[OSM-dev] Please, not another data structure discussion! (was: Contest for the best UI: Howto enter ways in the applet?)

[Immanuel Scholz]

Hi,

> I haven't studied these tools, but as far as I understand, the
> conventional GIS wisdom is to use a three level data model:
>
>  1. nodes with a latitude and longitude
>  2. lines made up of a sequence of connected nodes
>  3. polygons made up of a sequence of lines

The two real world data applications I saw (one mapping forests in Germany, 
Saxony and the other mapping streets in Germany, Bavaria) did not have a 
sequence of connected nodes.

Both had a line only span of exactly two nodes. One had the polygon consisting 
of a mixture of lines and nodes (which was very crappy) and the other a 
sequence of those two-node-lines.

Both had properties attached to every object (node, lines, polygons).

Only the forest stuff had support for areas at all.


However, GPX uses the above mentioned scheme, but GPX does not fit in OSM (as 
told hundrets of times before).


> Today OSM has a two level data model that doesn't support polygons
> (3) and our lines (2) can only be single line segments made up of
> two nodes.

This is only parital true. osm DOES have ways, it is only not implemented in 
the server. JOSM had support for ways from the first version.
The data layout of ways are already defined. What we need is the applet (and 
the server) to have it implemented. What we don't need is just another 
discussion about the intern data structure.

Sorry for my rude tone, but I am very tired of discussing AGAIN stuff we 
already discussed month ago.


> Today we can have attributes only for line segments, 
> not for longer runs or chains of segments.

Completly wrong. Ways have properties, nodes have properties...


> The lines can be roads, rivers, coast lines, outlines of forests,
> or administrative borders.  Country A can be a polygon surrounded
> by border lines L1, L2, L3 and country B can be a polygon
> surrounded by lines L4, L5, and L2.  Thus L2 is the border line
> between countries A and B.  Storing the line L2 once, and being
> able to reuse it for polygons A and B is a major argument for this
> data model.

Finding some use cases is always a luring argument to make things more 
complex.

I am fine with multinode line segments too (in fact, it was my first 
understanding of the data layout until I got corrected by Steve). But I am 
fine with the more simple two-node-only model of lines too. Since the latter 
is implemented and quite fits our needs, I don't see any reason to switch to 
something more complex.


> I have no prior GIS experience, so I don't know how traditional
> mapping user interfaces are designed.

Heterogen! The dream of "the one data model" is a fake. This just does not 
exist and this is good. Vector data and gis mapping is just to dependend on 
the application to have one scheme that rule them all (My opinion).


Ciao, Imi.