[mkgmap-dev] Fix Sea Patch

Mon Jan 4 19:40:57 GMT 2021

Hi Ticker,

I use coastline data that is generated from just line data and the algorithm used is not able to properly determine whether polygons are islands or water. It does however make sure polygons are closed, so I don't have problems with having to join ways together and hence I don't use close-gaps or extend sea-sectors. 

I'm unclear exactly what an anti-island is supposed to be - I have considered it to be a separate bit of sea, and in my data I have lots of those that are correct. They are only incorrect if they are within the sea, rather than the land. For example, if there is a coast road with sea on both sides of it, I am likely to have a separate sea polygon within the land if its connection to the sea is under a bridge. The patch checks for sea within sea and land within land. It doesn't then go back and recursively check for nested problems - I suspect this would be quite an overhead. 

The code is certainly not perfect, but does produce enormously improved results for me at least. I can still see at least one error in my map, but I think the remaining errors are tile boundary issues that are resolved differently on adjacent tiles. I am thinking of a small app to read the mkgmap log file and the coastline data, reversing whatever mkgmap says is wrong. By running mkgmap without splitter on just the coastline data I should be able to avoid any tile boundary problems.

Cheers,
Mike

-----Original Message-----
From: Ticker Berkin [mailto:rwb-mkgmap at jagit.co.uk] 
Sent: 04 January 2021 13:58
To: Development list for mkgmap <mkgmap-dev at lists.mkgmap.org.uk>
Subject: Re: [mkgmap-dev] Fix Sea Patch

Hi Mike & Gerd

I always try to use the coastline data in the downloaded OSM map to
generate the sea with option:
  --generate-sea=multipolygon,extend-sea
-sectors,close-gaps=750
and have only ever had 1 problem with faulty
coastline. In this case it wasn't a reversed section but a lake had
been tagged with coastline, worse, in the wrong direction so it wasn't
treated as an internal sea, rather an island, flipping the rest of the
tile to sea.

I sometimes have problems with the continuation of the downloaded country's coastlines being cut off and ending up near the middle of tiles, but closer to the wrong edge or, when extended to the edge, they are in the same direction as an adjacent coastline, because the un-crossing is outside the downloaded area.

@Mike - Do you use --coastlinefile to circumvent this type of problem and is there an advantage to this over using --precomp-sea? The data seems to need a lot of correcting!

If arbitrary bits of coastline are in the wrong direction, should you have another phase, after joining in the same direction and removing anti/islands. Try to join against their direction and moving closed ways to an ambiguous closed list and other ways that joined to an ambiguous linear list. Then see if the ambiguous ones can be resolved, maybe by weighting by the direction of the majority.

I feel that some of your fixes of land/sea are too late in the flow of code, ie after generateSeaBackgound has been set.

It seems a reasonably assertion that all anti-islands are a mistake in the data. I think the only one is the Caspian Sea, which is unlikely to be fully contained in a tile.

checkIslands(), to do the full job, would need to look at the nesting of sea/land/sea... so, given above, I don't think it worthwhile.

In verifyHits(), I don't see where any changes that resolves "adjacent hits in same direction" are applied. The 'fix' could make it totally wrong rather than partially wrong. Looking for 3 in the same direction and reversing the middle one might be better.

As I said earlier, I've never had a problem where this type of fixing would be of any use.

The swapping between land and sea needs to do more than just re-tagging; the multi-polygon handling needs to be considered and also the fullArea of sea.

Ticker

On Sat, 2021-01-02 at 10:32 +0000, Gerd Petermann wrote:
> Hi Mike,
> 
> > This was removed some time ago because it didn't work properly
> Do you have a version number where this happened? I only know the
> floodblocker option which is still evaluated in the source.
> 
> Gerd
> 
> 
> ________________________________________
> Von: mkgmap-dev <mkgmap-dev-bounces at lists.mkgmap.org.uk> im Auftrag
> von Mike Baggaley <mike at tvage.co.uk>
> Gesendet: Donnerstag, 31. Dezember 2020 16:17
> An: 'Development list for mkgmap'
> Betreff: [mkgmap-dev] Fix Sea Patch
> 
> Hi Gerd,
> 
> Mkgmap used to contain some code that attempted to correct incorrect
> sea and
> land data where polygons were ordered in the wrong direction. This
> was
> removed some time ago because it didn't work properly. I have put
> together
> the attached patch that works much better and have been using it for
> some
> time. With the sea data that I use, which contains quite a large
> number of
> incorrectly ordered polygons, it manages to correct every error it
> finds. It
> won't be able to fix really gross errors, but providing most of the
> data is
> correct it will produce good results.
> 
> Please review and commit if happy.
> 
> Cheers,
> Mike
> _______________________________________________
> mkgmap-dev mailing list
> mkgmap-dev at lists.mkgmap.org.uk
> http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev