feat: generalize multimodal link chooser

[sebhoerl]

Currently, the access/egress link of a car trip is only chosen based on the origin/destination facilities of the trip. We have use cases, where we know the location of a car (at a P&R station), so the "access link" should not be the activity location, but rather the location of the vehicle / station. The generalization of the interface as proposed in this PR allows implementing this kind of custom behavior (for instance by passing additional information through the Attributes of the RoutingRequest).

[Janekdererste]

Do we need the other decideOnLink method then?

[sebhoerl]

@Janekdererste This is just in the default implementation, because this one doesn't care about access/egress, but only about the facility location. So decideAccessLink and decideEgressLink delegate to decideOnLink, otherwise we duplicate the same code. But yes, we can make this one private in the default implementation.

[Janekdererste]

Sorry, @sebhoerl, I didn't read carefully enough. I thought the method is still in the interface. You implemented it just like I would have expected :-)

[d-roeder]

I really like the idea of ​​making this configurable.

However, when looking at the code I noticed one thing that might be problematic or lead to unintended behaviour. In

matsim-libs/matsim/src/main/java/org/matsim/core/router/NetworkRoutingInclAccessEgressModule.java

Line 206 in b487b41

Id<Link> endLinkId = toFacility.getLinkId();

and following the behavior of this PR is overridden when the facility is assigned to a link. This could possibly be mitigated if accessEgressModeToLink is used. Not sure. However, I think there should be no hidden magic when whe have the explicit LinkChooser.

Another point (whith a scope somehow outside of this PR) is that in

matsim-libs/matsim/src/main/java/org/matsim/core/router/NetworkRoutingInclAccessEgressModule.java

Line 191 in b487b41

Coord startCoord = NetworkUtils.findNearestPointOnLink(toFacility.getCoord(),egressActLink);

nearestPointOnLink is used. I think this leads to wormholes unless relativePositionOnLink is used, as the agent walks to any point on the link, but de facto starts at the end.

[sebhoerl]

Hi @d-roeder, I just took a quick look at the code. The toFacility is actually the destination of the car trip, and egressActLink is the link chosen by the link chooser. So I think everything is fine, or am I mistaken?

[d-roeder]

oh, you're right 👀

It could just be a bit confusing if the toFacility has no link set and NetworkRoutingInclAccessEgressModule then assumes that startLink==endLink. This makes sense in the default logic, but if arbitrary links can be set, then this is no longer necessarily correct.