[Talk-GB] Non-intuitive addresses

Sarah Hoffmann lonvia at denofr.de
Mon Feb 14 17:13:08 UTC 2022


On Sun, Feb 13, 2022 at 09:53:17PM +0000, Tom Crocker wrote:
> In particular, the wiki says, inter alia, that
> >
> > "When using addr:place=*, make sure to keep the name consistent with the
> > place=* object which the house number refers to."
> >
> > If those instructions are adhered to, then it follows that addr:place
> > isn't suitable for addresses where the street level line of the address
> > isn't a recognised place name.
> >
> 
> My understanding is that until 2018 the wiki said you did not need a
> matching place=* but text to that effect was introduced by Dieterdreist
> when he discovered that nominatim required there to be a matching object,
> which he misunderstood as place=* and that wasn't corrected.[1] I don't
> think it's documented what nominatim will allow to match for addr:place but
> I think it also allows landuse, some but not all place (not locality, ?
> others), and not building. Perhaps Sarah Hoffmann can enlighten us. So the
> addition was made so addr:place worked with nominatim but was inaccurate.
> This was recently strengthened by Mateusz Konieczny with regards to it
> being about keeping data consistent. Sarah (Lonvia) reintroduced the point
> that it need not match a place=* and now we have a somewhat mangled
> sentence that leaves us in no-mans land. However, it is also why I add the
> relevant area of named landuse and mentioned the idea of introducing a
> place=building_complex (or something) tag.

Nominatim indeed did require for a long time that a place with the same
name to exist. This had nothing to do with the definition of addr:place
but had internal reasons. Nominatim does not keep a full entry
internally for each housenumber but searches for a "parent place" and
reuses the information from there. This would occasionally lead to
unexpected behaviour where the tagging was okay but Nominatim still
wouldn't find the address. So people indeed added the hint in the wiki
that there must be some place object.

I've fixed most of the short-comings related to the parent place last
year. Nominatim will now happily accept any name in addr:place. If it
finds a place with the same name, it will still use it as a parent
place (very useful when your parent place comes with translations or
alternative names) but if there is no such place, then it'll work as
well. With Nominatim now working as expected, I've removed the sentence
in the wiki that was there for the benefit of Nominatim.

That said I'd still consider it generally a good idea to have a place
with the name used in addr:place, may that be a place node, a landuse
area, a building or whatever. For one thing, it gives the next
mapper a clue what the address refers to. That's good for quality
assurance. If there is a building or residential area named 'Abbots
Walk' plus addr:place=Abbots Walk, it is immediately obvious that the
address is correct. The other useful outcome is that the name shows up
on maps and can be found on its own (without the housenumber).

There are rare cases where the part in addr:place is really only used
for the address or there is no real-world object that can be said to
take the name. That's where you shouldn't add an artifical place just
for the geocoder.

Sarah



More information about the Talk-GB mailing list