Hi Thorsten The OutOfMemoryError: Java heap space exceptions are always in the MultiPolygonRelation processing when creating a 2 matrices of BitSets to work out the relationship between inner areas. At a minimum it needs sqr(numAreas) * 2 / 8 bytes and when there are 399992 areas (tile sea_2949120_360448.osm.pbf) this would is more than 37GiB. There are a few other tiles requiring 435, 35, 44 & 37 GiB. By this time all my threads have crashed.
Taking the first example (2949120_360448), there are 10352 land polygons that intersect with the sea tile area and, after these are clipped to the tile there are 17073 areas of land (it is expected that clipping will increase the number of areas). However merging these areas results in 399991 land areas. This should be impossible but I think the standard java area utilities can do this if a coastline is "twisted" on itself. I've been using planet-260706.osm.pbf/osmcoastline[_filter] and, today, land- polygons-split with data from 16-Jul-26 Ticker On Sun, 2026-07-12 at 13:03 +0200, Thorsten Kukuk wrote: > > Hi Ticker, > > on my machine (AMD Ryzen 9 PRO with 64GB RAM) it takes 6 minutes if > everything is Ok. But with my own self generated shapefile, I don't > remember the times of my tests with the upstream ones. I think they > where much longer. > I don't know exactly how long it took with my old machine (was a 8 year > old cheap Intel machine with 32GB RAM). It was always quick, too, but I > don't remember the exact times. But must be far below an hour, else I > wouldn't have generated the sea data daily. > My own shapefiles have always splitted polygons, since this is the > default of osmcoastline. > > I will try to make my self generated shape file available the next days. > The file from > July, 8th has a huge amount of wrong coastline ways. Newer files should > be better, but looking at my latest "osmcoastline" run, there are still > 66 wrong rings and somebody managed again to break the coastline :( > > Regards, > Thorsten > _______________________________________________ mkgmap-dev mailing list -- [email protected] To unsubscribe send an email to [email protected] %(web_page_url)slistinfo/%(_internal_name)s
