logo separator

[mkgmap-dev] Gaps in surfaces

From Ticker Berkin rwb-mkgmap at jagit.co.uk on Fri Oct 6 15:19:30 BST 2023

Hi Gerd

I've now looked at your example (mp-gap2.osm). With my toMapUnit/toHighPrec patch the problem
almost disappears. I don't see any significant difference when I also apply your mp-cut-coord-
pool.patch.

I can't see a way of fixing these gaps/overlaps without disabling a lot of filter point-removal,
maybe by setting Coord.preserved() and ensuring added points for polygon cutting are on both the
inner and outer polygon, which I thought they were, with the algorithm cutting through the
complete structure and then bits being merged later if possible.

Ticker

On Thu, 2023-10-05 at 16:37 +0100, Ticker Berkin wrote:
> Hi Gerd
> 
> I've changed the way mapUnit and highPrec Coords are calculated slightly so that the exact 1/2
> unit boundary values are consistent between negative and positive. This gets rid of my couple
> of
> abnormalities with a -32 delta. The old method added or subtracted 1/2 to make ((int)double)
> do
> rounding, which assigns the exact 1/2 to the larger abs value.
> 
> toHighPrec(), as in yesterdays change, keeps its bounds consistent with the mapUnit it is
> splitting up but is now coded in a similar way to toMapUnit.
> 
> I think rounding to get MapUnits is a mistake but changing it now could have drastic
> consequences, it should have been floorToNegInfinity, giving the nice cyclic scale -128..+127
> <<
> 24. As it is, we get -128..+128 << 24. 
> 
> Equator & Grenwhich Meridian behave well, but I've always had doubts about behaviour around
> 128
> 
> Hope this makes sense - patch attached
> 
> Sorry, haven't thought about the actual problem with the gaps yet.
> 
> Ticker
> 
> On Thu, 2023-10-05 at 07:30 +0000, Gerd Petermann wrote:
> > Hi Ticker,
> > 
> > reg. the change in toHighPrec() :  I must confess that I never doubted why positive values
> > should be rounded differently to negative values.
> > This is also done in Utils.toMapUnit() and I just adapted that code. If you still want to
> > change that I think Math.rint() would get you closer to the original implementation.
> > Question is if or why this difference in rounding is/was needed. There should be visible
> > effects near equator and at Greenwhich or at 180°.
> > 
> > Anyhow, I don't think this is the solution for the gap problem which is caused by further
> > conversions from highprec int to double and back.
> > An alternative solution for that problem would be the attached patch which makes use of the
> > coord pool.
> > No idea which one has more effects on performance.
> > 
> > Gerd
> > 
> > 
> > 
> > 
> > ________________________________________
> > Von: mkgmap-dev <mkgmap-dev-bounces at lists.mkgmap.org.uk> im Auftrag von Ticker Berkin
> > <rwb-mkgmap at jagit.co.uk>
> > Gesendet: Mittwoch, 4. Oktober 2023 15:57
> > An: Development list for mkgmap
> > Betreff: Re: [mkgmap-dev] Gaps in surfaces
> > 
> > Hi Gerd
> > 
> > I've been thinking about how to stop the small ranges of decimal degrees from generating
> > MapUnit/delta=+32 and MapUnit+1/delta=-32.
> > 
> > Without changing the MapUnit value, but truncating the highPrec calc, eg, in Coord.java
> > 
> >         private static int toHighPrec(double degrees) {
> >                 final double CVT_HP = ((double)(1 << HIGH_PREC_BITS)) / 360;
> >                 return (int)Math.floor(degrees * CVT_HP);
> >         }
> > 
> > this problem almost goes away, with deltas between -31 .. +32 except for 2 instances of
> > delta
> > of
> > -32 I get while building the Britain-and-ireland.osm.pbf. I need to work out why these are
> > happening.
> > 
> > Although it is unnatural to have this range rather than -32..+31, it doesn't matter. It
> > probably
> > could be fixed by using Math.ceil instead or reversing where the delta goes.
> > getAlternativePositions() will generated delta values approx -48..+48.
> > 
> > I don't get failures from LineClipperTest with this change.
> > I do get failures from GmapsuppTest, SimpleRouteTest & SortTest, but I think these are not
> > significant to this issue.
> > 
> > As far as the gap between areas is concerned, I haven't looked at this yet but I'll see if
> > my
> > change has similar effects to your patch to Coord.
> > 
> > Ticker
> > 
> > On Wed, 2023-10-04 at 08:16 +0000, Gerd Petermann wrote:
> > > Hi Ticker,
> > > 
> > > I've uploaded the test case: https://files.mkgmap.org.uk/detail/564
> > > I had to modify the data a bit since the default style doesn't render natural=heath
> > > 
> > > Gerd
> > > 
> > > ________________________________________
> > > Von: mkgmap-dev <mkgmap-dev-bounces at lists.mkgmap.org.uk> im Auftrag von Gerd Petermann
> > > <gpetermann_muenchen at hotmail.com>
> > > Gesendet: Montag, 2. Oktober 2023 16:58
> > > An: Development list for mkgmap
> > > Betreff: Re: [mkgmap-dev] Gaps in surfaces
> > > 
> > > Hi Ticker,
> > > 
> > > the word overflow might be confusing. The problem is that we want to use only 6 bits for
> > > the
> > > delta, but we store values from -32 .. 32.
> > > The special case with Michael example is that one coord with such an extreme delta is used
> > > (after converting to double) in Area.subtract() and the returned coord is converted back
> > > but
> > > get's the other extreme. In the end, only the 24 bit value is written to the map, and that
> > > causes the gap.
> > > 
> > > 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, 2. Oktober 2023 15:55
> > > An: Development list for mkgmap
> > > Betreff: Re: [mkgmap-dev] Gaps in surfaces
> > > 
> > > Hi Gerd
> > > 
> > > Considering no_hp-overflow_alpha.patch:
> > > 
> > > It seems wrong to change a 24bit coordinate. Utils.toMapUnit() is well defined to do the
> > > expected rounding with the conversion.
> > > 
> > > The actual deltas are local to Coord.java and, apart from their use by
> > > getAlternativePositions,
> > > are just used to get back to the highPrec coord value.
> > > 
> > > The deltas are stored as byte, so the value of +32 causes no arithmetic problems
> > > generally.
> > > 
> > > getAlternativePositions(), however, should handle any complications with the +32, but it
> > > looks
> > > like it mostly does, bumping up or down the 24bit coord if the abs(deltas) are > 16. It it
> > > really possible that modLxxDelta can overflow a byte here?
> > > 
> > > Haven't looked at LineClipperTest yet.
> > > 
> > > Ticker
> > > 
> > > On Fri, 2023-09-29 at 06:57 +0000, Gerd Petermann wrote:
> > > > Hi all,
> > > > 
> > > > I've found out that this problem is caused by a flaw in the "high precision" code. Under
> > > > special conditions, two points with almost identical coordinates are internally
> > > > represented
> > > > with very different values. There is a possible overflow in the delta values, the value
> > > > +32
> > > > should not occur as it cannot be represented with 6 bits, but the calculations produce
> > > > this
> > > > value.
> > > > I think I have an ugly fix for this but the resulting map still shows (smaller) gaps and
> > > > a
> > > > unit test needs corrections, so there is more to do.
> > > > Attached is a patch that checks for this overflow. Work in progress and maybe causes
> > > > trouble
> > > > in other areas, e.g. in South America where we have negative lat/lon values.
> > > > 
> > > > @Ticker: The unit test für LineClipperTest fails. I am not sure if only the test has to
> > > > be
> > > > changed (how) or if the code in LineClipper can be improved. I seem to remember that you
> > > > suggested to use
> > > > the code in ShapeSplitter instead.
> > > > 
> > > 
> > > _______________________________________________
> > > 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
> > > _______________________________________________
> > > 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
> > _______________________________________________
> > 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



More information about the mkgmap-dev mailing list