Thilo Hannemann wrote:

Hi Mark,

Am 09.09.2009 um 19:58 schrieb Mark Burton:

I will modify the code that the merging is applied even earlier.
Shouldn't be too hard to do.

Did you do that? I would be keen to try a version because I notice on
my first few trips with the Nuvi that sometimes it says "keep
right/left" when it should say "turn right/left". I think that is
because the way looks like this:

Some remarks about the merge lines concept, and the reasoning why it will not help much and needs to be extended/changed.

1. There is absolutely no need to merge lines in case the way continues at 0° angle (no change in direction).
2. There is no time penalty in Mapsource for a way zig-zagging, as long as there is no junction/change of way. The more the angle approaches 180° , the bigger the time penalty.
2a 180° exact i.e. only change in direction of way in osm database without oneway attribute does not matter of course, On the other hand every single route viapoint you set additionally will give you a time penalty. So if you route along a completely straight line and you put additional via points, the calculated arrival time gets bigger. This of course is irrelevant to mkgmap. The effect of the time penalty for via points, is only around 5-20seconds, so should be irrelevant (maybe maximum 1-2% of added time) to total travel time of a route.
3. If two ways meet on a straight line:
a) no time penalty is added at all
b) no turn indication is given
This means that you can well turn around a whole circle without any turn indication and many way changes, as long as the junctions themselves lie on a straight line. There is not even an indication of streetname changes. So in the directions window of mapsource, or route window of GPS it will not tell you that you get on another street (meaning another streetname) as long as you continue straight.

Taking into account the above:
Yes a merging lines patch will help for bicycle/pedestrian routing. Though much more effective would be a patch that actually changes the angle streets meet.

Let's first look at what happens when two streets meet at extreme angle. notice total time is 03:27min (with no junction at all it would be 01:28):
original

Concept1
So if two ways meet:
       y
       y
       y
xxxxxxxx
       y
       y
       y

We should make additional ways like this (this will help a tiny bit, but not much). ("." means that 2 ways lie on top of each other, ";" means at this point there are four ways on top of each other, so the direction change is not at the junction but done on the intermediary way)
       y
       y
       .
xxx. . .xxxx
       .
       y
       y
concept1

So we notice now that the time is actually the same as if there was no junction 1:28min. Note, no junction/direction change is shown at all (the GPS will still warn however when approaching). This achieves the same effect as the merge line patch would do in case both streets have the same name.

Concept 2:
By having the turn on the intermediary way, instead of the junction itself) we completely avoid the time penalty (though we also miss the name change of the street in the route window, nor any indication in the route window about the turn). Therefore we have to change the above still a tiny bit. By playing around with changing the distance I have found out that we need to put nodes at least around 1m distance of each other (otherwise we get short arc problem), so while the Concept1 I chose to have a rather large distance for the straight part, I now reduce the distance to 1m between each node (this means that we need to have 2m additional width for the junction - Concept1 could be achieved on 1m), the even dirtier hack to get junctions back into the route window has to make the whole thing rounded. So assume you come from top (north) and want to continue to the right (east) we have to add a new way like this (only showing added ways, we should never remove the original ways of course, so that crossing straight is still possible). The new way is drawn as .
                            y
                            y
                          .
                          ..
                                             xxxxxxxxxxxxxxx
This distance    |               | being 5.4m, while distance from north to south should be made around 1m. By doing this we get the best possible. a) we are notified about the junction/street change, b) we are taking a time penalty for turning. This time penalty is however very small (5-10 seconds), and should be more or less equal to what we need in real life to orientate here and change streets.

The proof in Mapsource of concept 2 (time penalty decreased to a mere 12 seconds), reducing the distance between the two additional nodes to the junction of 1m makes this more or less invisible:
concept2

The much bigger advantage however is, that the above concept would allow us on GPS to always route along the maximum of 100 direction changes (at least the old GPS can't route over more than 100 direction changes without via point) and over more or less indefinite distances in Mapsource (think Spain to Germany using small streets, cycleways and so on only). Transferring such a long route to GPS is even possible, using WinGDB to insert a via point every 99th direction change (Mapsource puts direction change points internally, but does not show them, they are saved in the .gdb file which is created if you save a calculated route. This way you could theoretically transfer a route going along 99.000 junctions to your GPS which is still easily able to recalculate it (well I'm sure GPS runs out of memory long before, but we will be able to have as long routes as fit into the memory, I think routes over 1000-2000km should still work).

A closeup of the work shown in JOSM zoomed up very close looks like this (red is original way):
closeup josm

Is there anyone who thinks the above concept2 can be implemented into mkgmap? I don't have the skills to code this. But maybe someone has the skills, and feels motivated by knowing that shitty routing for bicycle/pedestrian use, can be easily adapted by changing junctions with the above concept. Without implementing the above, we will never have good routing in Mapsource or on GPS. Using concept2 would get splendid routing (and make the merge lines patch discussed before obsolete!). Off course the above would not be needed, if Garmin would make the penalty adjustable, but that day will probably not come......

Also as a benegit the keep left vs turn left could be solved this way (increasing the angle of the junction high enough to get turn instead of keep, or in contrast decrease angle of the road you want to stay on so that it reminds you to keep left/right).

See attached also screenshots of one more example (classic junction) and a osm testfile for you to play with.