[Talk-ca] GeoBase and Canvec import tools

Frank Steggink steggink at steggink.org
Fri Sep 11 03:56:53 BST 2009


Hi Richard,

> This approach makes for many more change sets and each change set is
> likely to be a single object of some sort.  They still have to break
> these up further if the single relation is too big, but they aim for a
> changeset of about 1,000 items total.
>   
Too bad that the NTS tiles usually contain more data, even in rural 
areas. But I can understand the reasoning. It keeps imports more 
manageable. Importing data is not just making sure the data gets on the 
server, but also verification, eventually clean up, making connections, etc.
> I'm finding frequent upload failures when dealing with changesets
> larger than about 16,000 nodes.  Because of the nature of these change
> sets, I can end up dropping thousands of unconnected nodes all over
> the place.  Messy.  ;-)
>   
Did you use the bulkload script? I had the same issue with my very first 
import attempt last week, and cleaned up the 7000 nodes by downloading 
the changeset, adjust the XML, and upload that with JOSM.

Since that, I've used JOSM, but uploading is very slow. On average it 
takes about 1 hr to upload 8000 changes, but so far it has only hung on 
me once. That happened last Saturday. Probably there was some 
maintenance going on.
> One other approach that the French community is taking is to use
> postGIS to detect objects that overlap similar objects that already
> exist.  This is analogous to roadMatcher and it works on more object
> types, directly in the (local) database.
>   
Sounds interesting. Did they mention how they would do it? When buffers 
are used for example, it is easy to exclude roads which happen to run 
parallel along existing ways... But anything better than Roadmatcher is 
actually a win ;)
> There promises to be more information coming from the French community
> and the rest of the import group in the next while.  I hope that those
> in the Canadian community, particularly those developing tools can
> consider these two approaches.
>   
Keep up informing us about it :)

Regards,

Frank





More information about the Talk-ca mailing list