On Wed, Nov 11, 2015 at 6:48 AM, David Earl <da...@frankieandshadow.com> wrote:
> Having been negative, now the opposite... > > Why stop at ref? > > There are lots of places where it would help to group things uniquely > rather than by a simple text string. Names are the obvious next one. If we > put the names on a separate shared object, you can then tell when they are > actually the same street, or whatever, rather than just a coincidence. > There are two entirely separate Love Lane in Cambridge, for example, and > many High Street all over the place. How do you know they are the same, > especially if they don't all interconnect, or conversely they do, but are > actually different. > Associated street relations <http://wiki.openstreetmap.org/wiki/Relation:associatedStreet> do this, but it appears there's a big of a micromapping curve to it. > Also you could deal with the case of things which have different names > rather than the bodges we have at present. For example, where a street has > different names on opposites sides which are rare, but do exist. Or where > you have a part of a street known by the name of its terrace of houses on > one section, but where the street is really continuous. Or a bridge has its > own name but the street with a name continues across it. And that's even > without starting on different languages. > > Which then begs the question: why not do all or most of the tags this way? > Share tags between multiple objects as a matter of course when those > objects collectively represent one thing. And different but overlapping > subsets of objects represent different things. Then maybe say if we're > doing it this way some of the time, why not do it all of the time, even for > single objects, just for consistency. > The reason I've singled out automobile routes specifically on this has been that mapping this on ways instead of relations are inconsistent with how other routes are presently handled (that is, with relations) and maintaining this information in this weird one-off exception is a maintainability nightmare that just isn't present with relations. Relations do present their own challenges, however, with road routes, these are relatively easy to edit and validate (compared to, say, bus routes (which, seemingly increasingly as available funding diminishes, tend to have a lot of repeat members which make setting the order very difficult) or administrative boundaries). > In which case, are relations just in the picture because they happen to > provide a crude mechanism for this which already exists? > I would argue that ways and routes are separate entities for which it is exceedingly rare to find a 1:1 congruence (go ahead and try finding a single member route=road relation for which there are no changes in lane patterns including turn lanes, turn restrictions, bridges, tunnels, or any other objects that would cause a way to be split, and therefore ref=* on relations is a crude mechanism for which we now have an elegant solution available. I can only think of one example of a route that has just one member offhand, and that's relation 3632355 <http://www.openstreetmap.org/relation/3632355>, OK 34B. It connects US 283 to the unincorporated village of Brinkman, OK in Greer County, and the highway itself is very likely to be decommissioned in the current round of budget cuts or around the time OklaDOT gets a pothole report and wonders why it's still in the state inventory. The town's bank closed in 1927 and half the town burned down in 1929, with the town disincorporated not long after. The school (which at one point had 450 students enrolled) closed in 1965, and the post office closed 5 years later, and the railroad stopped serving the location in 1972. Judging by the two dwellings and lack of vehicles in aerial photography, I would hazard to guess that the village is within a rounding error of 0 population today. Almost any route in literally almost anywhere more relevant than Brinkman is very likely to have a route comprising of more than one way. I'm aware of other possible one-member routes, however, these are all in similarly dead-since-the-1930s western Oklahoma/Kansas locations that are unlikely to stay in state inventory once OklaDOT or KDOT realizes that it's control points for it haven't had any significant population, if any permanent population at all, in roughly 80 years. If all tags were done this way might we get to a point where we are > representing real world objects as a whole rather than trying to infer it > from groups of tags which happen to be similar and are split up purely for > convenience of managing other things like bus routes or bridges which run > along part of the real object. I'm not clear what you're getting at here, but if I'm reading that right (and I don't honestly believe I am), you're objecting to relations as a data type, because the whole of a complex object/site/route/whatever can be inferred from other entities? If that's the case, inference from groups of tags on what would otherwise be member objects is NP-complete (specifically, the clique problem, except on a massive scale). Relations essentially spell it out essentially eliminating that problem, which is why relations were introduced in the first place as the third data type.
_______________________________________________ Tagging mailing list Tagging@openstreetmap.org https://lists.openstreetmap.org/listinfo/tagging
