[mkgmap-dev] low-res-opt branch: error message from ShapeSplitter for self-intersecting multipolygon

Mon Jun 14 18:11:25 BST 2021

Hi Gerd

Looking at the first few GBR problems, it is shapeMergeFilter trying to
merge lots of adjacent, irregular, fields. They are all simple
polygons. It seems to be generating backward and forward lines all over
the place that don't make much sense, mostly not having a hole at the
end of them.

So I think ShapeSplitter_11 hasn't been proved faulty yet. I'll look at
shapeMerge.

Ticker

On Mon, 2021-06-14 at 17:14 +0100, Ticker Berkin wrote:
> But going back to running GBR, I'm getting more errors - I'll have to
> see if there is a compromise.
> 
> Ticker
> 
> 
> On Mon, 2021-06-14 at 16:39 +0100, Ticker Berkin wrote:
> > Hi Gerd
> > 
> > Made some progress! I needed to disable the EPS_HP=2 tolerance
> > along
> > the cut line otherwise there were correct cases that could be
> > interpreted incorrectly. I also needed not to chuck away very small
> > areas otherwise your test complained. The major problem was valid
> > shapes that had parts with zero area when cut and passed through
> > the
> > same point on the cut-line as another line.
> > 
> > The main tiles in xxx.zip now build without error with order-by and
> > no
> > -order-by, even with ShapeMergeFilter maxPoints = 65535.
> > 
> > However, the overview builder, when --no-order-by and --improve
> > -overview, fails splitting. The shape presented to ShapeSplitter
> > has
> > obvious self-intersections (with --order-by, shape merge is
> > disabled
> > in
> > the overview builder)
> > 
> > Ticker
> > 
> > On Mon, 2021-06-14 at 08:28 +0000, Gerd Petermann wrote:
> > > Hi Ticker,
> > > 
> > > I am still testing with the files in 
> > > https://files.mkgmap.org.uk/download/507/xxx.zip
> > > 
> > > I applied the attached patch to enable massive merging and
> > > simulating
> > > possible splits and your splitShapeFix_10_lowRes.patch and I see
> > > several error messages.
> > > The previously attached example comes from xxx\63240004.osm.pbf
> > > 
> > > My options:
> > > --output-dir=e:\ld  --gmapi --generate
> > > -sea=multipolygon,floodblocker 
> > > --preserve-element-order --order-by-decreasing-area
> > > --allow-reverse-merge   --improve-overview --min-size-polygon=2
> > > -c e:\xxx\template.args
> > > 
> > > Gerd
> > > 
> > > ________________________________________
> > > Von: mkgmap-dev <mkgmap-dev-bounces at lists.mkgmap.org.uk> im
> > > Auftrag
> > > von Ticker Berkin <rwb-mkgmap at jagit.co.uk>
> > > Gesendet: Montag, 14. Juni 2021 10:19
> > > An: Development list for mkgmap
> > > Betreff: Re: [mkgmap-dev] low-res-opt branch: error message from
> > > ShapeSplitter for self-intersecting multipolygon
> > > 
> > > Hi Gerd
> > > 
> > > Any form of self-intersection might cause problems to
> > > ShapeSplitter.
> > > Not sure what you mean by "connected hole visited.." - I presume
> > > something like the outer edge of the shape is clockwise, with a
> > > cut
> > > leading to a hole that is also clockwise; for this to happen the
> > > in/out
> > > lines from the edge to the hole conceptually cross.
> > > 
> > > I'm just running low-res-opt on BritishIsles with
> > > ShapeMergeFilter
> > > maxPoints=65535 to see if I can get a problem to work on. What is
> > > your
> > > test area?
> > > 
> > > Also, the latest development versions of the suspect code areas
> > > are
> > > in
> > > different branches. Should I attempt to have a composite with the
> > > faster-mp bits of MP handling and low-res-opt version of
> > > ShapeMergeFilter?
> > > 
> > > Your attached example looks OK and I wouldn't have expected it to
> > > cause
> > > any difficulties for the later versions of ShapeSplitter. A
> > > possible
> > > difficulty might be an almost horizontal line close to the in/out
> > > lines
> > > to the lower shape. I hope my attempt to allow a bit of tolerance
> > > where
> > > calculated points on the cut-line end up 1/2 HP units the wrong
> > > way
> > > round isn't causing other problems. Do you have this as an
> > > example
> > > I
> > > can test?
> > > 
> > > Ticker
> > > 
> > > 
> > > On Sun, 2021-06-13 at 14:17 +0000, Gerd Petermann wrote:
> > > > Hi Ticker,
> > > > 
> > > > OK, my current understanding is that ShapeSplitter may fail to
> > > > split
> > > > 8-like shapes or shapes where the connected hole is visited in
> > > > the
> > > > "wrong" direction. Those problem cases are possibly created by
> > > > 1) MultipolygonRelation.joinways()
> > > > 2) MultipolygonCutter
> > > > 3) ShapeMergeFilter (maybe)
> > > > 
> > > > Do you see a way to avoid them in those routines? Attached is
> > > > another
> > > > example that doesn't work.
> > > > 
> > > > Gerd
> > > 
> > > _______________________________________________
> > > mkgmap-dev mailing list
> > > mkgmap-dev at lists.mkgmap.org.uk
> > > https://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
> > > _______________________________________________
> > > mkgmap-dev mailing list
> > > mkgmap-dev at lists.mkgmap.org.uk
> > _______________________________________________
> > mkgmap-dev mailing list
> > mkgmap-dev at lists.mkgmap.org.uk
> > https://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
> _______________________________________________
> mkgmap-dev mailing list
> mkgmap-dev at lists.mkgmap.org.uk
> https://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev