[mkgmap-dev] Please test branch NET-no-NOD

Tue Oct 22 16:51:46 BST 2019

Hi Gerd

Testing mkgmap-NET-no-NOD-r4304 with 2 tile local area with option
--x-check-routing-island-len=700

Some build statistics:

SEVE: uk.me.parabola.imgfmt.app.net.RoadNetwork  74210002.osm.pbf:
check for routing islands found 293 islands
SEVE: uk.me.parabola.imgfmt.app.net.RoadNetwork  74210002.osm.pbf:
routing island check took 5252 ms
SEVE: uk.me.parabola.imgfmt.app.net.RoadNetwork  74210001.osm.pbf:
check for routing islands found 2820 islands
SEVE: uk.me.parabola.imgfmt.app.net.RoadNetwork  74210001.osm.pbf:
routing island check took 45755 ms

Total time taken: 1 minute 53 seconds
gmapsupp.img 13719552 bytes
block size: 2048
        .       20992
MAKEGMAP.MPS    138
00007421.MDR    1121108
000MYTYP.TYP    1580
74210001.RGN    3802657
74210001.TRE    17874
74210001.LBL    758810
74210001.NET    1536856
74210001.NOD    2556416
74210002.RGN    1645275
74210002.TRE    8820
74210002.LBL    268782
74210002.NET    718899
74210002.NOD    1241600
00007421.SRT    912

For comparison, testing with r4295 (I get similar figures for trunk+):

Total time taken: 1 minute 8 seconds
gmapsupp.img 14710784 bytes
Block size: 2048
        .       22016
MAKEGMAP.MPS    138
00007421.MDR    1121108
000MYTYP.TYP    1580
74210001.RGN    3802683
74210001.TRE    17874
74210001.LBL    758810
74210001.NET    1562592
74210001.NOD    3285327
74210002.RGN    1645275
74210002.TRE    8820
74210002.LBL    268782
74210002.NET    722291
74210002.NOD    1477491
00007421.SRT    912

So, for my test case, with my style, which probably creates quite a few
more isolated networks than the default style:

It adds quite a lot of processing time (increased by 66%)

Reduces the NOD size significantly (22%) and full gmapsupp.img by 6%.

It fixes a lot of "Route calculation errors" where the nearest road to
the start/end point is an isolated network. However the behaviour of
this changes in a way not apparent to the map user if what should have
been an isolated network overlaps tiles.

I vote for keeping something like this provided the routing-island
detection only happens of the length option is specified (needs a
better name than --x-check-routing-island-len)

The unconnected road handling should be moved out of StyledConverter
and handled as part of the length controlled general logic.

Ticker