Multipolygons and disappearing coastlines
Hi team, I have been asked by the Open Sea Map people to help them generate some maps with marine objects and while working on that I have noticed that if a way is part of a multipolygon then all its tags get removed. This is unfortunate because (in the test map I am looking at around Keil) the coastline is part of a multipolygon (some administrative boundary) and, therefore, the coastline doesn't appear in the map because its tags have been removed (or is it global warming?) So, is putting the coastline into a multipolygon a reasonable OSM thing? If so, should mkgmap be breaking the coastline by removing its tags? If not, how do we fix that? Any thoughts? Cheers, Mark
Mark Burton schrieb:
So, is putting the coastline into a multipolygon a reasonable OSM thing?
A coastline is often also a boundary and boundaries are multipolygons in OSM.
If so, should mkgmap be breaking the coastline by removing its tags? If not, how do we fix that?
My guess is, that the multipolygon handling in mkgmap is not working perfectly. See an older thread on this list where I suggest a workaround. (I check if the MP has no inner elements and change the type=multipolygon to type=boundarypolygon if this is the case). Chris
Hi Chris,
Mark Burton schrieb:
So, is putting the coastline into a multipolygon a reasonable OSM thing?
A coastline is often also a boundary and boundaries are multipolygons in OSM.
If so, should mkgmap be breaking the coastline by removing its tags? If not, how do we fix that?
My guess is, that the multipolygon handling in mkgmap is not working perfectly.
See an older thread on this list where I suggest a workaround. (I check if the MP has no inner elements and change the type=multipolygon to type=boundarypolygon if this is the case).
Chris
OK, that's understood. But why do we have to trash the original ways when we make a multipolygon from them? Would it not be better to give the relation a duplicate of the way to trash so that the original way can go on to serve its original purpose. Perhaps we should do this: diff --git a/src/uk/me/parabola/mkgmap/reader/osm/xml/Osm5XmlHandler.java b/src/uk/me/parabola/mkgmap/reader/osm/xml/Osm5XmlHandler.java index acc5cf9..fbb1d57 100644 --- a/src/uk/me/parabola/mkgmap/reader/osm/xml/Osm5XmlHandler.java +++ b/src/uk/me/parabola/mkgmap/reader/osm/xml/Osm5XmlHandler.java @@ -266,10 +266,12 @@ class Osm5XmlHandler extends DefaultHandler { private void startInRelation(String qName, Attributes attributes) { if (qName.equals("member")) { long id = idVal(attributes.getValue("ref")); - Element el; + Element el = null; String type = attributes.getValue("type"); - if ("way".equals(type)){ - el = wayMap.get(id); + if ("way".equals(type)) { + Way way = wayMap.get(id); + if(way != null) + el = way.duplicate(); } else if ("node".equals(type)) { el = nodeMap.get(id); if(el == null) { I just tried this patch and (just like magic) the coastline reappears. Cheers, Mark
Hi Chris,
Mark Burton schrieb:
So, is putting the coastline into a multipolygon a reasonable OSM thing?
A coastline is often also a boundary and boundaries are multipolygons in OSM.
If so, should mkgmap be breaking the coastline by removing its tags? If not, how do we fix that?
My guess is, that the multipolygon handling in mkgmap is not working perfectly.
See an older thread on this list where I suggest a workaround. (I check if the MP has no inner elements and change the type=multipolygon to type=boundarypolygon if this is the case).
Chris
OK, that's understood.
But why do we have to trash the original ways when we make a multipolygon from them? Would it not be better to give the relation a duplicate of the way to trash so that the original way can go on to serve its original purpose. Perhaps we should do this:
diff --git a/src/uk/me/parabola/mkgmap/reader/osm/xml/Osm5XmlHandler.java b/src/uk/me/parabola/mkgmap/reader/osm/xml/Osm5XmlHandler.java index acc5cf9..fbb1d57 100644 --- a/src/uk/me/parabola/mkgmap/reader/osm/xml/Osm5XmlHandler.java +++ b/src/uk/me/parabola/mkgmap/reader/osm/xml/Osm5XmlHandler.java @@ -266,10 +266,12 @@ class Osm5XmlHandler extends DefaultHandler { private void startInRelation(String qName, Attributes attributes) { if (qName.equals("member")) { long id = idVal(attributes.getValue("ref")); - Element el; + Element el = null; String type = attributes.getValue("type"); - if ("way".equals(type)){ - el = wayMap.get(id); + if ("way".equals(type)) { + Way way = wayMap.get(id); + if(way != null) + el = way.duplicate(); } else if ("node".equals(type)) { el = nodeMap.get(id); if(el == null) {
I just tried this patch and (just like magic) the coastline reappears.
Cheers,
Mark
Hi Mark, I didn't try your patch but the idea is good. It is also more compliant to the rules described in the OSM wiki (http://wiki.openstreetmap.org/wiki/Multipolygon#Tagging) than the current implementation. Mulitpolygon code should be reimplemented soon. As far as I am quite new to mkgmap I will need some time to do this. At the moment I see the following things to do: 1. Only simple polygons are supported. Lots of options described in http://wiki.openstreetmap.org/wiki/Multipolygon#Advanced_multipolygons are not implemented. I think the algorithm in the OSM wiki (http://wiki.openstreetmap.org/wiki/Relation:multipolygon/Algorithm) is a good starting point to make mkgmap implementation compatible. Some rules about tagging are also described there. 2. The PolygonSplitter destroys the polygons created by the current multipolygon code. Up to now I do not fully understand if thats a problem or not. But I would feel better if someone could replace the usage of the java.awt.Area class with some code that does not remove the mulitpolygon connections between outer and inner rings. 3. Up to now the MultiPolygonRelation class does support one outer ring only. This must be changed so that the MultiPolygonRelation class can create multiple polygons in the garmin format. I do not fully understand how the Garmin map is assembled from the mkgmap map format, so any hint is appreciated. WanMil
2. The PolygonSplitter destroys the polygons created by the current multipolygon code. Up to now I do not fully understand if thats a problem or not. But I would feel better if someone could replace the usage of the java.awt.Area class with some code that does not remove the mulitpolygon connections between outer and inner rings.
yes this is a problem. polygons across multiple tiles are broken. It's not possible to close a polygon by a reliable algorithm. instead splitter should keep all members of a polygon and mkgmap can clip at tile boundary. Have no Java knowledge but can provide a small example to demonstrate it.
On 26/12/09 22:11, Mark Burton wrote:
Hi Chris,
Mark Burton schrieb:
So, is putting the coastline into a multipolygon a reasonable OSM thing?
A coastline is often also a boundary and boundaries are multipolygons in OSM.
If so, should mkgmap be breaking the coastline by removing its tags? If not, how do we fix that?
My guess is, that the multipolygon handling in mkgmap is not working perfectly.
See an older thread on this list where I suggest a workaround. (I check if the MP has no inner elements and change the type=multipolygon to type=boundarypolygon if this is the case).
Chris
OK, that's understood.
But why do we have to trash the original ways when we make a multipolygon from them? Would it not be better to give the relation a duplicate of the way to trash so that the original way can go on to serve its original purpose. Perhaps we should do this:
diff --git a/src/uk/me/parabola/mkgmap/reader/osm/xml/Osm5XmlHandler.java b/src/uk/me/parabola/mkgmap/reader/osm/xml/Osm5XmlHandler.java index acc5cf9..fbb1d57 100644 --- a/src/uk/me/parabola/mkgmap/reader/osm/xml/Osm5XmlHandler.java +++ b/src/uk/me/parabola/mkgmap/reader/osm/xml/Osm5XmlHandler.java @@ -266,10 +266,12 @@ class Osm5XmlHandler extends DefaultHandler { private void startInRelation(String qName, Attributes attributes) { if (qName.equals("member")) { long id = idVal(attributes.getValue("ref")); - Element el; + Element el = null; String type = attributes.getValue("type"); - if ("way".equals(type)){ - el = wayMap.get(id); + if ("way".equals(type)) { + Way way = wayMap.get(id); + if(way != null) + el = way.duplicate(); } else if ("node".equals(type)) { el = nodeMap.get(id); if(el == null) {
I just tried this patch and (just like magic) the coastline reappears.
But do you also magically get an extra unwanted polygon when the relation is a cascading multipolygon? Aside from that, the multipolygon code already does duplicate the way before removing the tags from the original. Duplicating it again can't be the right solution. ..Steve
On 28/12/09 11:46, WanMil wrote: Hi
Aside from that, the multipolygon code already does duplicate the way before removing the tags from the original. Duplicating it again can't be the right solution.
Only the outer way is duplicated. Inner ways seems to be kept but all tags are removed.
Only tags from the inner way that are also present on the outer way are removed according to the code. But yes, maybe they should be duplicated as well, I'm not against duplicating the ways (and I like Mark's second patch), just against doing it twice without a rationale. This history as far as I can remember for removing tags from the ways in the manner that it is done is to prevent duplicate polygons in cases when people have needlessly/incorrectly/historically tagged both the relation and the ways that make it up, particularly in the lake-on-an-island-in-a-forest case. I've read the links that you posted earlier in the thread and I'd would like to see the code updated to follow the given algorithm now that there are a lot more relations being used. ..Steve
participants (5)
-
Apollinaris Schoell -
Chris-Hein Lunkhusen -
Mark Burton -
Steve Ratcliffe -
WanMil