[OSM-talk] Postmortem analysys

[Nic Roets]

On Fri, Jan 7, 2011 at 11:49 PM, Mike N. <niceman at att.net> wrote:
>> Mike, please don't blame the bot. Ungluing a node an just leaving it
>> there, is really looking for trouble. Some routing engine(s) glue
>> nodes together that are less than a few centimeters from each other.
>> Now you may want to complain that those routing engine(s) are buggy,
>> but that "bug" has historically made things easier rather than more
>> difficult. And going forward, I expect it to continue to be a
>> "feature" rather than a bug.
>
>  Consider me firmly in the "it's a bug" camp.   Routers in general work with
> data from different sources; but it's a bug in OSM to have an intended
> connection only be close but not connected.    There's no minimum node
> distance for disconnected nodes - just best practices to minimize database
> clutter from dupe nodes.  QA tools like Keepright make it feasible to
> monitor and maintain large areas in a fully correct topology.
>
> Do routing engines glue nodes from different layers?

Yes, provided they are close enough (for the engine(s) in question).

>  Do they automatically
> connect crossing ways on the same layer?

Only in the rare case where both crossing ways contain nodes at the
crossing point. (for the engine(s) in question).

Look, I'm saying that I do my part fix things that may cause problems
somewhere in the software stack. For example, I discovered that the
Appalachian trail was at some point the largest object in the DB.
Potlach and several other pieces of software would either bomb out or
take ages to respond. So I messaged the last editor and he agreed that
many of the nodes are redundant. So I deleted those nodes.

After ungluing a node, move one of them just a little bit. (Unless you
used a DGPS with a 10cm resolution and found that the centerlines are
in fact on top of each other). If you leave them on top of each other,
it's going to waste someone's time later on (either after a bot edit,
a keepright warning, a routing error or a user editing the area who is
completely oblivious to the possibility that two nodes can be on top
of each other).

A very simple way of reducing the problem inside the router will be to
move nodes by small random amounts, but I have more urgent bugs and
feature requests.

> Either modification would change
> the calculated route.
>
>
>
> _______________________________________________
> talk mailing list
> talk at openstreetmap.org
> http://lists.openstreetmap.org/listinfo/talk
>