Some points not linked to streets

[abyrd]

Alex has reported holes in a regional analysis of Switzerland. These appear in both regional accessibility analyses and single-point analyses.

His analyses are with the car mode. In the regional analyses, these are probably errors linking the origin to the network. In the single-point analyses they are probably errors linking destinations.

Note that the holes all seem to be in places where cells might be linked to a motorway. I have seen similar things in the Netherlands, north of Amsterdam where the new metro line runs in the middle of a highway.

[abyrd]

Alex has confirmed that the accessibility value at all these places is zero.

Not all unconnected areas are near motorways though. if connections to motorways are not allowed, they should still be connected to something. I suspect it’s finding the motorway, realizing it’s a motorway, and then not re-trying to find a nearby surface street. The same thing could be happening in the city center: it’s in car mode, and linking to a bicycle path or pedestrian path, realizing it’s not appropriate for cars, and then not retrying.

[abyrd]

Looking at com.conveyal.r5.analyst.TravelTimeComputer, only walking can work correctly 100% of the time, because we're linking to the street network before setting the routing mode. I'm tempted to make the default mode null so it will fail if we do this anywhere.

[abyrd]

After making the obvious fix and telling the origin linker what mode will be used, things get much better. But there are still some holes at the origin and destination ends. Another regional analysis with the new worker: Note that there are still tiny holes around some roads, and strangely an area of low accessibility around the highway. The latter may be due to directionality issues (cells linked to only one direction of the highway).

The first part may have something to do with dead ends or service roads. I can reproduce these holes around service roads, and very interestingly the holes appear even when I drop the origin inside the hole (of course it may just be linking to a point outside the hole, and the hole is still inaccessible).

In Portland:

[abyrd]

This has much improved, but we need to make some decisions here about how to deal with ways that have different permissions in opposite directions (including one-way streets and big boulevards). We can link to a way if it's traversable in either direction, or only if it's traversable in both directions. And we probably want to exclude motorways, linking only to surface streets, since the linked points generally represent businesses or homes.

[ansoncfit]

The idea of linking to multiple nearby streets could mitigate the originally reported issue here. Linking to multiple streets was discussed in (#228) and came to mind again looking into #505.