[OSM-talk] full history of a way?

[MP]

>  If you often have this kind of query, consider downloading and suitably
> preparing some old planet files for quick access to old data with proper
> referential integrity.

I don't think this is possible for me, as having several old planet
files in state where it could be quickly searched (so probably
unpacked) would probably need several hundred GB of disk space. Which
is too much for me.

Luckily I need this type of query only rarely.

> > Perhaps this could be improved in 0.7, by having some
> > "[way|relation]/#id/history-full" call, which will return
> history of
> > referenced way (or relation), including history of all node (ways)
> > that was referenced at least in one of the historic or current
> > versions. Sort of combination of /history and /full
> >
>
>  I'd rather hope someone builds some kind of "time machine" service that has
> these capabilities (plus a "give me this area as it was last year" API
> call). There is no pressing reason to burden the central API with that once
> we figure out how to publish full history dumps.

There are no full history dumps currently - having such dump would
enable this type of query quite easily. I think the daily diffs could
be used for help if the changes are fresh (less than 14 days). Are
there any nearby plans for full history dumps? This shouldn't probably
be very hard, instead of having only one node/way/relation in the
result you will have many with same ID and different timestamp as the
history progresses. You'll have just to mark deleted objects somehow
(those that have history, but they are not in current data)

Basically, once the initial dump will be done, you can just
concatenate any changes :)

Well, there could be slight problem with pre-0.6 data (no order in
relations) and pre-0.5 (extra segment elements in addition to nodes
and ways), so this maybe need some conversions. I am not sure what was
before 0.4 (could be some problems too...)

So, I guess I'll have to write some tool to query the server for the
node histories then, as it probably is currently the most effective
way considering there won't be many queries...

Martin