[OSM-talk] Mapnik tileset coherency issues

Patrick Weber p.weber at ucl.ac.uk
Thu Oct 18 13:28:52 BST 2007



Steve Chilton wrote:
> Patrick raises an issue that many in the project often forget about, and
> that is the image that is presented to allcomers, whether familiar with
> the ins-and-outs of the project or not.
> A commercial project would background render a new tileset and release
> it in one go on a planned basis. This is not a commercial project and we
> have arranged (decided on?) a response mechanism that allows work to
> show as soon as possible - people just want to see their mapped stuff
> ASAP. This is what I mean by it being an inwardly focussed rather than
> outwardly focussed project in many ways.
>   
steve thanks for understanding and better expressing what my thoughts were.
> I do have qualms about seeing adjacent mapnik layer tiles rendered with
> seemingly different data. It can of course occur both in adjacent tiles
> on the same layer, and between the same area/tile on different layers.
>   
> Could I suggest that Jon (and any others) working on revising the
> process consider trying to apply some "adjacency" intelligence? So, for
> instance when rendering a dirty tile is it possible to check adjacent
> tiles (immediately there will be 4 abutting tiles at the 4 points of the
> compass, and 4 more which meet at the corners) for "dirtiness" and
> render them next? Equally, when rendering a particular dirty tile is it
> possible to immediately queue all iterations of that tile through the
> zoom levels?
>   
Yes both issues were what I was aiming at, to address adjaceny in both 
x,y terms, and z (zoom level) terms ? What I mean is, say a tile x,y,z 
is dirty and needs to be rerendered. You would include adjacent tiles on 
a given z , but also on lower z levels (say a dirty tile x,y at zoom 
level 12, would also mean all tiles contained by tile x,y at subsequent 
zoom levels 13,14 ... would be considered dirty in the same job). Could 
you render this batch of tiles than into a cache, before committing the 
whole batch to the live tileset, thus ensuring more consistency ?

Presumably, more tiles would get marked as dirty and be rerendered than 
at the moment, because consistency across x,y,z is considered ?
-------------- next part --------------
A non-text attachment was scrubbed...
Name: p.weber.vcf
Type: text/x-vcard
Size: 282 bytes
Desc: not available
URL: <http://lists.openstreetmap.org/pipermail/talk/attachments/20071018/12e7e1e0/attachment.vcf>


More information about the talk mailing list