Expose the change / change:lanes tags, resolved to direction of travel

[dsaderholm]

Following on from #127, and shaped the way you described there: resolved to the direction of travel, folded into one output, no per-direction fields.

Would you consider carrying OSM's change / change:lanes and exposing the value for our direction as a single output?

It is the closest thing OSM has to "may I change lanes here". change=no is a solid line. change:lanes=not_left|yes says the leftmost lane may not cross left while the next one may. overtaking is the European tag for roughly this and reads 0% in Utah, so I had assumed the concept was unmapped here until I enumerated the tags instead of guessing at them.

Measured on US-6 in Utah, 497 ways, 134 carrying it, grouped by how many lanes run in the forward direction:

1 forward lane    19   14%
2 forward lanes   57   43%
3 forward lanes   38   28%
4 forward lanes   14   10%

86% is on multi-lane sections, which surprised me. I had expected a two-lane-road tag. The values there are not about oncoming at all: change=no with two same-direction lanes is a work zone, a bridge, or a gore area, and that is a restriction nothing on the vehicle can see.

Coverage is lower on interstate, 17 of 400 I-15 motorway ways, and those encode the express lane barrier: no|not_left|yes|yes|yes|not_right|no, lane 1 closed and lane 2 unable to enter it.

For a lane-change feature this is more directly useful than the lane counts I asked for in #127, since it states the restriction rather than making us infer it from geometry.

Happy to test on hardware, same as before.

[dsaderholm]

Correcting my own numbers before you spend any time on this.

I measured one road. Checking a second changes the picture a lot:

US-6    134 of  497 ways   27%
I-15     17 of  400 ways    4%
US-89     0 of 1594 ways    0%

US-89 is a comparable road to US-6 in the same state, also with 2+1 passing
sections, and it carries the tag on nothing at all. So change coverage here is
one corridor rather than a general property of US roads, and the 86% multi-lane
split in my original post describes that single corridor.

That is weaker than I made it sound, and you should weigh it accordingly. The
argument that stands is the one about what the tag means when it is present:
change=no across two same-direction lanes is a restriction nothing on the
vehicle can perceive, and no other tag carries it. The argument that does not
stand is that it is widely available.

Entirely reasonable to close this as not worth the effort. I would rather you
had the real number than my first one.

[pfeiferj]

For this one I'll have to take a look through the osm wiki to see all the ways lanes are used. I think I've seen some other lane designations that are more common, so if there is a more common thing that gets more or less the same information I'd prefer that. Other than that, I'm not inherently opposed to this

[dsaderholm]

I went and counted rather than leave you to dig through the wiki. Worldwide object counts from taginfo, today:

overtaking             151,095
change:lanes            80,151
change:lanes:forward    52,650
change:lanes:backward   51,022
change                  11,105
median                   9,617
divider                  8,862
separation:left          7,528

Two things in there worth your time, and the first one favors your instinct over my request.

overtaking is 13.6x more common than what I asked for, and it answers a different question: it is the no-passing zone on a two-lane road, so it says whether you may cross into oncoming traffic. change says whether you may move to the next lane going your way. Both are useful and neither substitutes for the other. If only one of them is ever worth the work, overtaking is the better value, and it is the one I would drop this request for.

The second is that I asked for the wrong spelling. change:lanes is 7.2x more common than bare change, so the per-lane form is where the data actually is. If you do implement this, that is the one to read.

Nothing in the divider, median or separation family clears 10k, so there is no third option hiding in there.

Worth adding to the coverage numbers in my previous comment: overtaking is on 0% of all three Utah highways I sampled, so both of these tags look considerably more European than American. That is an argument for treating either as best effort rather than something a consumer can lean on, which is how we use every map field anyway.

[dsaderholm]

Retitled to name change:lanes as well, so the issue asks for the form the data is actually in rather than the rare one I opened it with. No change to the request itself, and the ordering I gave above stands: overtaking first if you are picking one.