Hi Gerd, Mike Considering --generate-sea=... and SeaGenerator.java. Following changes around r4381, many more tiles have land/sea flipped, and, looking at the log files, MultiPolygonRelation gives errors for the sea relation, complaining about inner polygons being in inner polygons etc. The code relating to extracting polygons from the coastline doesn't seem to understand what it is doing regarding the change in behaviour between land polygons on a sea tile and vice-versa and I have little confidence in its ability to handle complex coastlines. So, I've re-written the guts of it to fix these shortcomings. I haven't touched the "floodblocker" code, which I think should be decommissioned. Patch attached. There are a few other aspects that to this code that I think could be improved. Ticker On Thu, 2020-01-30 at 11:35 +0000, Ticker Berkin wrote:
Hi
Having just generated full britain-and-ireland with current trunk (r4432), I'm also now seeing some tiles (5 out of 101) with sea/land flipped when using option: --generate-sea="multipolygon,extend-sea-sectors,close-gaps=350" but when I process some of the same tiles with r4295 they are OK.
It's not a problem for me at the moment, I've simply replaced above with --precomp-sea=sea-latest.zip
However I prefer to use --generate-sea for various reasons: - saves downloading sea.zip every now and again. - island cut-outs match land features exactly, whereas there were slight differences with sea.zip. - not had these problems before. - no noticeable performance problems.
Sometime I'll try and pin down when the change happened. I hadn't noticed before because the tiles in the small map I use day-to-day don't show the problem.
Ticker
On Wed, 2020-01-29 at 22:50 +0000, Mike Baggaley wrote:
Hi Gerd, up to now I have built my UK map with the sea and land in a single pass using --generate-sea. However, some tiles are getting the sea and land inverted, typically where a very small amount of sea is in the tile. I was trying to see whether if I precompiled the sea, the problem would go away - I was assuming the tiles of precompiled sea would be bigger because they contained no other data, so the probability of inversion would be reduced. I was therefore first trying to generate precompiled sea, then use it to build my UK map. It may be that my assumption is incorrect and that it won't make any difference anyway.
Regards, Mike