Natick, MA

[harling]

I'm sure each of the individual symptoms can be addressed, whether by geometry, segment naming or what have you. In the end, though, I think the two-way ramp makes for a more resilient map in these cases. (Fewer segments, fewer junctions, more tolerant of handset GPS, less dependent on geometry.)

[jondrush]

Fewer segments? Seriously? Nearly every ramp diverges at one or both ends. So a two-way joined to 2 one-way is three segments. Two one-ways from start to finish is two segments. Plus you've taken away all sorts of decisions and penalties that the routing engine has to apply.

That is the crux of the difference between ramps and roads. Ramps almost always diverge. I've done lots of ramps each way. Editing is so much simpler without two-way ramps. I don't need to address 'symptoms', they just flat-out work, with virtually no URs. Why are we even debating this?

[banished]

Why are we even debating this?

I think because the road that got us to a reasonable definition of split vs. unsplit has been long and fraught with potholes -- like every other definition/agreement we've reached, eh? I would not discard it so easily. Both options work, but I urge caution establishing deviations to the wiki; else, the next *M who has actually read the wiki is going to change it back, if possible.

If it's a good idea that has most CMs supporting it, then it deserves to be documented in the wiki.

Best,
banished

[harling]

Nearly every ramp diverges at one or both ends. So a two-way joined to 2 one-way is three segments. Two one-ways from start to finish is two segments...

That is true. And when they diverge at both ends, as in the example that I posted earlier, there are six segments whether you use one-ways or two-ways. So you are correct that using two-way ramps sometimes results in one additional segment. (The only time it reduces the count by one is when you have a single two-way segment, which is rare.)

What is reduced is the total length of roadway on the map: replacing two very close, parallel stretches of road with one. That means one segment to adjust to get the geometry right instead of two; one segment for a sometimes flaky GPS to snap to instead of two.

Plus you've taken away all sorts of decisions and penalties that the routing engine has to apply.

Such as? Two-way segments maintain independent speeds for each direction, and as I understand it, junctions keep track of the cost of each segment transition separately.

That is the crux of the difference between ramps and roads. Ramps almost always diverge. I've done lots of ramps each way. Editing is so much simpler without two-way ramps. I don't need to address 'symptoms', they just flat-out work, with virtually no URs.

And I find the map a lot simpler to follow and easier to maintain, in some cases, with two-way ramps.

Why are we even debating this?

1) The guidelines (rightly) leave room for a certain amount of discretion on the part of the editor;
2) The AM primarily responsible for this area for the past couple years sometimes finds it preferable to do it this way; and
3) It has solved URs and MPs.

[WeeeZer14]

My main reason for avoiding two way ramps is that each direction usually has very different names (e.g. "Exit 123: blah blah" vs. "to highway blah N"). A two way segment should be unnamed so it picks up the proper one-way part that it (hopefully) connects to. But unnamed segments often get an eager editor trying to assign a name then you get confusing instructions.

That said, I am sure I have left some two way ramps in place where I have edited.

[jondrush]

So you are correct that using two-way ramps sometimes results in one additional segment. (The only time it reduces the count by one is when you have a single two-way segment, which is rare.)

Very common on inverted cloverleaf interchanges

What is reduced is the total length of roadway on the map: replacing two very close, parallel stretches of road with one. That means one segment to adjust to get the geometry right instead of two; one segment for a sometimes flaky GPS to snap to instead of two.

Plus you've taken away all sorts of decisions and penalties that the routing engine has to apply.

Such as? Two-way segments maintain independent speeds for each direction, and as I understand it, junctions keep track of the cost of each segment transition separately.

I don't care about total length of road on map, why would anyone? I only care about segment count, junction complexity and naming complexity. Two-way ramps are typically the simplest regions of the ramp, straights and simple bends. When we had a splitting tool I would adjust geometry once, then split. I've made a request of Waze to see if we can somewhat duplicate this functionality. Why do we care about flaky GPS on close ramps? There is only one solution to a ramp, you get on, you get off. Unlike split roads.

Every junction is a calculation that the routing server has to make. One-way to two one way is one decision. Two-way adds three decisions, at least. Also more ways for the editor to get it incorrect.

[harling]

I don't care about total length of road on map, why would anyone? I only care about segment count, junction complexity and naming complexity. Two-way ramps are typically the simplest regions of the ramp, straights and simple bends... Why do we care about flaky GPS on close ramps? There is only one solution to a ramp, you get on, you get off...

The "why" is that you get many of the same problems that arise when a road is too narrow to justify splitting it:

• visual clutter of parallel segments that are only a few feet apart
• the need to adjust the geometry of two segments instead of one
• recalculations when it looks like you are headed back onto the highway, resulting in URs
• GPS tracks that look like you're headed the wrong way on a one-way segment, resulting in MPs.

When we had a splitting tool I would adjust geometry once, then split. I've made a request of Waze to see if we can somewhat duplicate this functionality.

Considering how widely it was misused in Cartouche, I don't think making a "split two-way road" feature available to the WME world would be advisable, unless they figure out a way to undo it just as easily.

Every junction is a calculation that the routing server has to make. One-way to two one way is one decision. Two-way adds three decisions, at least.

The only difference in the example I provided, as far as the routing server is concerned, is the junction where the off- and on-ramps for Route 128 meet the two-way segment. Approaching that junction in either direction, only one turn is possible. So it adds the one-time fixed cost of one additional junction look-up; it does not add a new branch to the routing tree.

Also more ways for the editor to get it incorrect.

We can spend all day weighing the relative merits of storing an extra junction vs. storing two parallel sets of geometry nodes, of the possibility of an editor overlooking a turn restriction vs. the accuracy of smartphone GPS. The bottom line is, responsibility for this corner of the map tends to land on my shoulders; and I believe mapping it this way--which I find easy, reliable and effective, or I would have done it differently--is within the discretion allowed by the editing guidelines.