Hi Frederik,
On Sun, Nov 7, 2010 at 10:44 PM, Frederik Ramm <frede...@remote.org> wrote:
> Hi,
>
> I'm a major user of the --bp task in Osmosis, for the nightly Geofabrik
> country extracts. I cannot use the complete-something flags as that would
> take far too much time, and on the whole the "clip-incomplete-entities" mode
> seems not to be too bad for users.
>
> There's one thing that has popped up on this list however ("lost relations
> while clipping") which is an increasing concern to our users and which I
> hope I can fix. The problem is that with cascading relations, Osmosis will
> only include a relation if one of its children has been included already, so
> if
>
> node N1 is inside the clipping area,
> relation R1 contains node N1,
> relation R2 contains relation R1
>
> then relation R1 will alway be included in the output, but relation R2 will
> only be included if it comes after R1 in the input file.
>
> Now while using the temporary storage for nodes and ways would be too
> expensive for the nightly Geofabrik build, I think that temporarily storing
> relations would be acceptable (as there aren't too many). I'm thinking of
> adding a "cascadingRelations=yes" flag to the --bp/--bb tasks that would
>
> * process relations not containing other relations as usual,
> * throw all other relations onto a stack and process them in a suitable
> order afterwards
>
> I am aware that determining a "suitable order" could be difficult if you
> have many levels of cascasing (or even circular) relations, and that the
> output file would not be ordered properly, but I think such an option would
> mean a big improvement at a relatively low cost.
>
> 1. Any comments?
>
It sounds good to me.
> 2. Anyone done something like that already?
>
Not exactly, although it should be very similar to the existing completeXXX
functionality.
> 3. Is any of Osmosis' internal storage mechanisms suitable for buffering
> relations in this way or do I have to invent something new?
>
I'd try to use the existing completeXXX options as a starting point. The
existing file system stores should work for your purposes. If their
performance is not good enough then it might be necessary to investigate
other options. In that case if you did come up with a higher performance
design it might also be able to be used for the completeXXX options as well.
Cheers,
Brett
_______________________________________________
osmosis-dev mailing list
osmosis-dev@openstreetmap.org
http://lists.openstreetmap.org/listinfo/osmosis-dev
