In London, at least, it seems to be becoming common to put low cost businesses at the back of high strreet businesses such that they are only accessible by means of an unadopted** service road at the back of the high street. They will tend to have an address on the corresponding high street, as the service road has no name.
I’m thinking of one case I just discovered. Although I haven’t added an address, as I’m wary about adding an address when a Google search was used, even though not Google Maps, I know that the address is on the high street, but the only way of accessing the POI is to walk a total of almost 200 metres, almost half of which is along an unlit service road, with no footway, and significant amounts of fly-tipping - a route I’'d hope a router would normally disccourage. I actually only discovered it because there was a notice on the business that it back onto, telling you how to reach it.
Flats above shops may also only be reachable by such service roads, even though numbered on the main road.
Whlist one can use associatedStreet to associate a POI with street contained in its postal address, is there any recommended way of hinting that a POI is physically reached via some other way. I don’t consider detailed micro-mapping a practical solution, and even that may not work unless you invent phantom side footpaths, which are generally considered to break the rule about mapping what is on the ground.
** Unadopted refers to a highway, etc. that is not maintained at public expense.
I don’t get the problem. For a router, it doesn’t matter that the POI has a different address than the closest street. Take e.g. https://graphhopper.com/maps/?point=Eikenstraat%2C%202840%2C%20Rumst%2C%20Belgium&point=Print%26Bottle%2C%202840%2C%20Rumst%2C%20Belgium , it follows a service road without name at the end.
The problem you describe might occur in Nominatim, as that program always takes the nearest street to compose the address of a POI, even when there is an addr:street tag on the POI.But I consider that as a program bug, not as incorrect data
Well, to evaluate such cases, can you give us some examples for those POIs via an OSM permalink?
And what routing provider and what geo search engines have you tried so far to find out any strange routing?
See http://wiki.openstreetmap.org/wiki/Routing and http://wiki.openstreetmap.org/wiki/Search_engines
Sorry, got distracted onto other things.
The POI in question is https://www.openstreetmap.org/node/3054589815
Of the routers that are available from the main page, MapQuest (two instances) produces an impossible route, because it seems to go for the closest place on a significant road, which iis actually a long way from the real route and doesn’t even take you to front of the building (where there is a notice in the window, directing you to the right place, and the others get it right. OpenRouteService.org completely denied the existence of the private back alley that needs to be used to access it, and simply refused to route to anywhere on it, even if you positioned the destination on the way itself.
The reason that I would hope that a router would avoid the use of the back alley is that if I didn’t already know the location and had arrived, particularly by public transport, at night, and saw a fly-tipped private alley, I would most likely assume that I had been misrouted and it was unsafe to continue. I’d therefore only want that route to be generated if there was a strong hint that it was ther right route even though it looked unsafe. I suspect that MapQuest is also denying the existence of the alley.
Incidentally, the routers that do find a route for cycles and cars, also get it wrong, as they generate a U-turn at the end of a section of dual carriageway. That isn’t signed as no U-turn, so it not absolutely illegal to do so, but if you did so at any time except the early hours of the morning and were seen by the police, you would probably get a ticking off. Whilst I could hint this, it would not reflect what was actually on the ground.
For the case when the proper access road is physically far from the POI, and routing application might choose some other road, there has been a proposal on the Polish forum:
Relation with tags
This way, one can indicate (in explicit, machine-readable form) what’s the proper access road to a given place, in case where routers would choose automatically an unsuitable way.