Drop undesired tags in default.style about osm2pgsql HOT 6 CLOSED

Here's my style, feel free to grab: https://github.com/Komzpa/furry-sansa/blob/master/assets/osm2pgsql.style

Also it may be useful to have some other lines pulled from there, like area:highway -> phstore.

from osm2pgsql.

right now https://github.com/openstreetmap/osm2pgsql/blob/4802a525fd/output-pgsql.c#L162 will error if phstore is used without hstore mode. Perhaps we should change that so we can add area:highway and other tags to default.style?

Probably a longer-term issue

from osm2pgsql.

Adding more of those "delete"s to the default style probably does make sense. On the other hand, does anyone have an estimate of how much disk space is saved by not including them in the hstore? Is it worth the change and the potential loss in backwards compatibility?

from osm2pgsql.

I doubt it would break backwards compatibility in any practical way. It is unlikely that anyone is using flags like tiger:upload_uuid except for the db on errol. People doing very unusual things with osm2pgsql should not use default.style from the package for consuming minutely updates or else we would never be able to change it.

It's also worth noting that for many of these tags they are generally only ever dropped in minutely diffs because most editors will remove them.

from osm2pgsql.

On the other hand, does anyone have an estimate of how much disk space is saved by not including them in the hstore?

The tags hstore column for the linestring table on errol is 17GB. Assuming most ways don't have enough tags to be TOASTed (reasonable, unless there's a lot of tags in which case they're probably droppable) here are some numbers

There's 13.5m tiger ways. With the droppable tiger tags, a rough estimate puts the count of 100 million tags dropped, average 15 byte key. If you assume 30 bytes total per tag that's 3GB, for a reduction of about 18% from TIGER alone. The tags index is about twice the size, and it'd reduce that too.

Yahoo is another big set, with about 40 million tags. They're not 8-bit characters, but it's a few percent.

Overall I'd guess 25%-50% of the tags hstore could be gotten of rid of.

from osm2pgsql.

Fixed in 1331b6c

from osm2pgsql.