[OSM-talk] About large streets

[Simon Hewison]

Joerg Ostertag wrote:
> This way you'd have to add two tracks for all normal 2-way streets, but you 

Thinking back to my dim and distant past, when I write a route planning 
program on a Commodore 64, I modeled each node to have a number of links 
to surrounding junctions (I didn't bother with intermediate points 
curves, but stored the distance in the node). I did have to remember to 
specify both directions for every link, but it did work, and it never 
went the wrong way down a one-way street. However, it still couldn't 
deal with no-right-turn and no-left-turn.

My algorithm then wasn't brilliant, it consisted of first of all heading 
in the right direction, and finding any path it could (it had problems 
with rivers, and ended up backing up and starting from the previous 
node).. Once it had found a route, it proceeded to attempt to find 
better routes between portions of the route it had found, in an 
iterative manner. (It was a variant of Dijkstra's algorithm, but I 
hadn't heard of Dijkstra at the time)

That was in the days of a 1MHz cpu, and relatively little RAM. These 
days, most PC-based route planning software seems to do things 
differently. I found a useful link.

http://ai-depot.com/BotNavigation/Path.html

So, when once of us (possibly me) starts a route-planning bit of 
software using OSM data, it would first of all need to get a data 
snapshot of an bounding box a little bigger than the start and end 
points, and then build it's own data structure, knowing that if there's 
a segment from A->B then unless there's tags otherwise, B->A is also 
possible. Therefore the OSM database doesn't itself need to store 
segments from A->B and B->A because that could be left to the route 
planning engine's own data structure.

-- 
Simon Hewison