ratrun

@jansoe: Your above referenced branch tree/CitySpeedMerge returns a now "404 Not found". I wanted to take a look on what you did in this branch. Could you provide this work...

I believe that the elevation data is just not precise enough such that a computed slope can reasonably influence the weighting and that therefore the results will be rather random....

With the extreme example above I wanted to visualise the problem with the approach. I expect there high values for the computed slope values, where nobody would expect that even...

I checked the result for Germany: `bicycleRouteWayMembers.size(): 26117 superRouteRouteMembers.size(): 1377` Do you think this is worth a new configuration parameter such that the default (e.g. for car) is to not...

Now I implemented the releasing of the memory after pass1: [715c7d4](https://github.com/graphhopper/graphhopper/pull/2516/commits/715c7d453b8bc4d175b7db80348a7575b69c2f7d).

@karussell Please provide feeback on this one, it looks to get lost. What do you want me to change or is there no chance that this gets merged?

@easbar Thanks for the feedback. > What would really help merging this PR is some statistics about the number of relations, collection sizes, memory usage and import time when using...

Thanks for checking. Here is one prominent real world example: The EV9 in Austria for this route: https://graphhopper.com/maps/?point=47.552028%2C16.090164&point=47.511694%2C16.042957&locale=de&elevation=true&profile=bike&use_miles=false&selected_detail=Elevation&layer=TF%20Cycle . If you look more closely, I'm sure you will find lots...

> As this is now following the official bike route ... I think the route you meant was [this](https://graphhopper.com/maps/?point=47.552028%2C16.090164&point=47.531748%2C16.071882&point=47.511694%2C16.042957&profile=bike&layer=TF%20Cycle). Yes, exactly. Please note that somebody changed back the EV9 from...

>If we would introduce e.g. area routing that require more complex relations parsing or a third pass or similar, then I would lean towards of accepting this complexity as this...