[Talk-GB] addr:place cleanup process

[Mark Goodge]

On 16/01/2022 11:09, Cj Malone wrote:

> Not advocating this for OSM, but as an aside, in an unrelated dataset I
> completely gave up on structured addresses. I simply use streetAddress,
> locality, region, postcode, country as freeform text fields. (I pretty
> sure I copied the idea from another spec I came across)
> 
> streetAddress holds all the complexity, it can be as as simple as
> "housenumber roadname", or packed with unit housename, multiple roads,
> places, whatever. It can't really be parsed by a machine, but is great
> to display to end users. locality are basically region line 2 and 3 on
> a letter, it's not enforced to be actual admin levels.

FWIW, the PAF uses a similar concept, even though it's administered 
differently. It does have a large set of possible fields, most of which 
will be null for the majority of addresses (and hardly any address will 
have data in all of them). But the fields are divided into three logical 
groups: premises, street and locality, each of which has multiple 
permutations within it, together with the required fields of post town 
and postcode which are something of an add-on to the first three.

Most commercial applications which store addresses don't enforce any 
structure apart from the postcode/zipcode and, if applicable, the 
country. British software tends to have a field for post town, while US 
software generally has a field for state (and internationalises it as 
county for British addresses, something which is technically wrong but 
doesn't break anything), but other than that the rest of it is generally 
just Address line 1, Address line 2, etc.

Mark