Hi! First tests are good :). But something strange happens with location of some house numbers. For example Pulawska 67 (in Warszawa city) is here: http://osm.org/go/0OyvXjAp5 but on Mapsource is shown here: http://osm.org/go/0OyvSy29r This is more than 1 kilometer apart! Also addresses like Municypalna 14 are shifted about 50m. Another bug is related to uncommon house numbers like 12a, 67/69, etc. When given an address like Municypalna 12a, Mapsource tries to locate Municypalna 12 (which is elsewhere, probably interpolated because no node with such housenumber is in OSM). When I enter address Pulawska 67/69, Mapsource shows Pulawska 623. Main branch of mkgmap behaves in similar way, but addresses are shifted slightly different (When entering Municypalna 12a, Mapsource searches for Municypalna 23, Pulawska 67 is in different place - still wrong). This doesn't seem to be related with --code-page switch. best regards Michal Rogala 2014-02-17 10:39 GMT+01:00 Michał Rogala <michal.rogala@gmail.com>:

thanks!

I will test it :)

best regards

Michal Rogala

2014-02-16 22:01 GMT+01:00 Steve Ratcliffe <steve@parabola.me.uk>:

On 05/02/14 12:10, Michał Rogala wrote:

...

...

map without code-page 1250 (default setting) has incomplete index - I can't search for almost any street (only a few available, no housenumber search).

Hi,

I've created the 'sort' branch to work on this. If you want to try out the latest you can download mkgmap-sort-r3056. I created a map of Poland with just --route --index and it seems OK. But you probably know better what to look for.

..Steve

http://www.mkgmap.org.uk/download/mkgmap-sort-r3056.zip

_______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev

On 19/02/14 09:04, Michał Rogala wrote:

I've started tests on GPSMap 62s and I'm not able to search any city in the index ;(.

Is that with a gmapsupp created with mkgmap or via MapSource? You could also try using code-page=1252 which would have fairly similar result for names that are originally in 1250. Best wishes ..Steve

On 21/02/14 10:27, Michał Rogala wrote:

Unfortunately code-page 1252 lacks some 1250 characters that are crucial in polish language ;(

Well, so does the default ascii... In both cases letters that don't exist are transliterated. Is that not true? The results should be no worse for 1252 than the default, or could be made so. ..Steve

On 21/02/14 15:35, Michał Rogala wrote:

Oh, I see your point - using 1252 as a replacement to default encoding which has broken address search? In that case it's an idea worth trying :)

Yes, that was the idea. Not that I am giving up on making it work with the default sort if it is at all possible. Was it just cities that could not be found? If you do use mapsource/basecamp to load the map onto the device does it work then? ..Steve

On 26/02/14 14:09, Michał Rogala wrote:

Map uploaded using Mapsource works fine, gmapsupp.img created by mkgmap has broken address search. On my device you need to specify country and city first - I can't get past this stage so I can't tell if street name search works.

I'm afraid I cannot reproduce this on my etrex 30. I do have random problems with a map that is built without a --bounds file, but this doesn't seem to be any better whether the gmapsupp is created with mkgmap or mapsource. When built with a bound file I seem to get a good selection of cities and streets. Although the etrex 30 allows you to search all cities, you can specify one and all seems to work OK. Of course without local knowledge I might be missing a lot of streets, but from your message I gather that you couldn't find anything at all. I've uploaded the file I produced as I would be interested if it works for you at all. I used the latest verion of mkgmap-sort, although it doesn't have many changes since the version you might have used before. ..Steve

hi! Address search in your map works very well on my device. So maybe it's related to your changes in code or problem lies within content of the index (maybe some characters in other city names are improperly transliterated). Could you send me your mkgmap version? BTW - is there any tool to debug those indexes? best regards Michal Rogala 2014-02-27 22:59 GMT+01:00 Steve Ratcliffe <steve@parabola.me.uk>:

On 27/02/14 21:57, Steve Ratcliffe wrote:

...

I've uploaded the file

... and it is here: http://files.mkgmap.org.uk/download/181/gmapsupp.img

..Steve _______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev

I used mkgmap-sort-r3056 - I'll test your version. I'm using my own bounds files - they were always working fine. best regards Michal Rogala 2014-03-03 12:53 GMT+01:00 Steve Ratcliffe <steve@parabola.me.uk>:

Hi Michał

Address search in your map works very well on my device. So maybe it's

...

related to your changes in code or problem lies within content of the index (maybe some characters in other city names are improperly transliterated). Could you send me your mkgmap version?

Its http://www.mkgmap.org.uk/download/mkgmap-sort-r3067.zip

There were no changes that were not in that.

But I did use a bounds file eg: http://osm2.pleiades.uni- wuppertal.de/bounds/20140128/bounds_20140128.zip which is important to get good search coverage.

BTW - is there any tool to debug those indexes?

...

There is test.display.MdrDisplay and test.check.MdrCheck in the display project at http://svn.mkgmap.org.uk/display/trunk But there are many false warning, so it is not that useful unless you know what you are looking for.

..Steve

_______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev

Hi! Here are files I use to generate my maps: custom pbf extract (500 MB): https://mega.co.nz/#!9kEWWSxJ!RKRgqEJiaJ_691G12q3Gc2fj9Ldg5FadMiG03q47S6w boundaries: https://mega.co.nz/#!VlUlWZDB!lK4BYN7mIn3VNIWMPT2EJ7jCcVK8FbDMIw3SXhf5ECQ best regards Michal Rogala 2014-03-04 19:44 GMT+01:00 Michał Rogala <michal.rogala@gmail.com>:

Hi!

Unfortunately - latest mkgmap-sort still generates maps that don't work on my device :(.So I think problem lies in index content - after entering first letter in search box, GPSMap 62 claims that it can't find any city. As far as I know there is no city in Poland that starts with diacritic letter. But my map covers also border parts of Germany, Ukraine, Russia, Czech Republic, Slovakia and Belarus - maybe other charsets characters are improperly transliterated.

Tomorrow I will send links to files that I use to compile my maps.

best regards

Michal Rogala

2014-03-03 19:47 GMT+01:00 Michał Rogala <michal.rogala@gmail.com>:

...

I used mkgmap-sort-r3056 - I'll test your version. I'm using my own bounds files - they were always working fine.

best regards

Michal Rogala

2014-03-03 12:53 GMT+01:00 Steve Ratcliffe <steve@parabola.me.uk>:

Hi Michał

...

Address search in your map works very well on my device. So maybe it's

...

related to your changes in code or problem lies within content of the index (maybe some characters in other city names are improperly transliterated). Could you send me your mkgmap version?

Its http://www.mkgmap.org.uk/download/mkgmap-sort-r3067.zip

There were no changes that were not in that.

But I did use a bounds file eg: http://osm2.pleiades.uni- wuppertal.de/bounds/20140128/bounds_20140128.zip which is important to get good search coverage.

BTW - is there any tool to debug those indexes?

...

There is test.display.MdrDisplay and test.check.MdrCheck in the display project at http://svn.mkgmap.org.uk/display/trunk But there are many false warning, so it is not that useful unless you know what you are looking for.

..Steve

_______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev

Hi

Did you have time to look for the cause of this bug?

Sorry I've been distracted by all the routing excitement :) I shall take a look later today. ..Steve

Hi Michał

Did you have time to look for the cause of this bug? :)

Just a quick update to say that I have looked at the problem, but no solution. I'm not really sure what I'm looking for of course :) I've found one thing that might be a problem which I will investigate at a later date. ..Steve

Hi Michał

I suppose problem may be in content of an index - your map of Poland works well on my device but any map compiled by me (using latest mkgmap release) using data I sent you (bigger Europe slice than boundary of Poland) has broken index. Yesterday I made some tests and neither compilation with latin1 not 1252 nor 1250 works. Index in MapSource works well, maps uploaded using MapSource also work well.

I've found a few differences with the gmapsupp that we produce and the attached patch fixes most of them. I would appreciate if you could test the current patch. Trouble is, that nothing that I've changed really appears to be specific to that part of the world and so I don't why it does not affect everyone. There are a few names that begin with spaces, which are possibly a problem. I would like to get rid of all initial spaces from names and I will attempt that in the next patch. Best wishes ..Steve

On 25/03/14 13:37, Michał Rogala wrote:

Thanks a lot. Could you send me binary file of patched mkgmap? I don't have any tools installed to patch and compile it.

I've just created a new branch which includes that patch so you can get it from: http://www.mkgmap.org.uk/download/mkgmap-sort2-r3125.zip ..Steve

On 25/03/14 16:34, Michał Rogala wrote:

Thanks. Unfortunately maps generated using this version still don't work ;(. So the only explanation at this time is that the bug is caused by names starting with spaces.

You say that it works when loaded on to the device using MapSource? In that case it cannot be the spaces themselves that is a problem, and I made a fix for dealing with them that was included in that patch. So how badly does it not work? Can you search for anything? Is it only addresses you can't find? Since a small map worked you may be able to find a group of tiles that work and work out if there is a particular tile that makes it go wrong. ..Steve

Hi! This bug is tricky as hell :). Here are my observations: 1) No separate tile causes broken index. This only happens when all (or at least unknown combination of) tiles are compiled together. 2) After many hours of testing I've selected 2 particular tiles. When I remove them from my compilation, index starts working again. 3) Those 2 tiles - 6600435 and 6600436 cover border parts of Ukraine. I suppose Ukrainian transliterated names cause problems. My guess would be cities like: Volia-Baraniets'Ka or Ostrozhiets' (note apostrophe at the end) 4) I uploaded my splitted map here: https://mega.co.nz/#!8l1jCLSR!x9FhsFVzLa6T7OfvYb-jYglP6voK04BcVwWHeeu7Hhw here are generated boundaries: https://mega.co.nz/#!R5k1iZqS!2UDuibgP1BAXNbh5FMMp3yt_5U5dyJBrh5hvObMVl0o If you try to compile this map, please use particular config file, as I may be crucial for replication of this bug: verbose max-jobs country-name=Polska name-tag-list=name:pl,name generate-sea location-autofill=is_in,nearest tdbfile index add-pois-to-areas link-pois-to-ways remove-short-arcs preserve-element-order country-abbr=POL route net gmapsupp nsis check-styles process-exits process-destination housenumbers overview-mapnumber=66007001 overview-levels=6:16,7:15 best regards Michal Rogala 2014-03-26 10:54 GMT+01:00 Michał Rogala <michal.rogala@gmail.com>:

2014-03-25 23:40 GMT+01:00 Steve Ratcliffe <steve@parabola.me.uk>:

...

You say that it works when loaded on to the device using MapSource? In that case it cannot be the spaces themselves that is a problem, and I made a fix for dealing with them that was included in that patch.

So how badly does it not work? Can you search for anything? Is it only addresses you can't find?

Since a small map worked you may be able to find a group of tiles that work and work out if there is a particular tile that makes it go wrong.

Maps work on MapSource and BaseCamp. Maps loaded using BaseCamp work on my device. To find an address Garmin asks you to enter country and then city (and later street). Device doesn't see any city to choose/search for.

I'll try to find tiles that cause this problem.

best regards

Michal Rogala

Hi

If you try to compile this map, please use particular config file, as I may be crucial for replication of this bug:

verbose max-jobs country-name=Polska name-tag-list=name:pl,name generate-sea location-autofill=is_in,nearest tdbfile index add-pois-to-areas link-pois-to-ways remove-short-arcs preserve-element-order country-abbr=POL route net gmapsupp nsis check-styles process-exits process-destination housenumbers overview-mapnumber=66007001 overview-levels=6:16,7:15

I noticed that this list does not inlucde a --bounds option. Can you confirm if you did also include the bounds file as an option? ..Steve

Hi Michał I believe that I have now found the problem thanks to your example. If I'm right, then the problem occurs when there are some cities that do not have a country assigned and also that the total number of cities is greater than 64k and the number of cities with a country is less than that. So it is mostly just unlucky based on the amount of data. With a perfect bounds file, then every city should have a country and the problem would not happen I think, so there was probably some significance about the tiles being at the edge in a different country. If not using a bounds file, then the problem is much more likely to happen and may explain why I often see problems with maps without bounds files. The fix is on the sort2 branch at r3163 or greater: http://www.mkgmap.org.uk/download/mkgmap-sort2-r3163.zip ..Steve

hi! Could you please merge this changes into main branch? So far I didn't notice any problems with this version of mkgmap. best regards Michal Rogala 2014-04-07 12:45 GMT+02:00 Michał Rogala <michal.rogala@gmail.com>:

Hi!

First tests indicate that your patch is working great :). Thanks a lot for fixing this bug :)

best regards

Michal Rogala

2014-04-02 17:50 GMT+02:00 Steve Ratcliffe <steve@parabola.me.uk>:

...

Hi Michał

I believe that I have now found the problem thanks to your example. If I'm right, then the problem occurs when there are some cities that do not have a country assigned and also that the total number of cities is greater than 64k and the number of cities with a country is less than that.

So it is mostly just unlucky based on the amount of data. With a perfect bounds file, then every city should have a country and the problem would not happen I think, so there was probably some significance about the tiles being at the edge in a different country.

If not using a bounds file, then the problem is much more likely to happen and may explain why I often see problems with maps without bounds files.

The fix is on the sort2 branch at r3163 or greater: http://www.mkgmap.org.uk/download/mkgmap-sort2-r3163.zip

..Steve _______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev

hi! Unfortunately, I've found an error with sort2 patch - many towns and villages are gone from index. This is not related to any map tile - it seems to be randomly distributed. MapSource/BaseCamp works fine. If you still have my map files - you can try to find towns like Mielnik, Czeremcha or Marysinek. best regards Michal Rogala 2014-04-20 17:10 GMT+02:00 Steve Ratcliffe <steve@parabola.me.uk>:

Hi

Could you please merge this changes into main branch? So far I didn't

...

notice any problems with this version of mkgmap.

OK great, thanks for testing.

@all: the sort2 branch only contains changes to make a gmapsupp more compatible, including a major problem although probably rearly seen, depending on the number of cities.

I plan to merge that and then create another branch for the unicode changes.

..Steve _______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev

Thanks for the hint with city search/address search difference - I've found a nasty bug in my style :). Sorry for your trouble. best regards Michal Rogala 2014-05-07 15:27 GMT+02:00 Steve Ratcliffe <steve@parabola.me.uk>:

On 05/05/14 13:45, Michał Rogala wrote:

...

Unfortunately, I've found an error with sort2 patch - many towns and villages are gone from index. This is not related to any map tile - it seems to be randomly distributed. MapSource/BaseCamp works fine.

If you still have my map files - you can try to find towns like Mielnik, Czeremcha or Marysinek.

I can find these in MapSource if I use city search, but they do not show up in the city field of address search.

I believe this is because those cities do not contain any named streets.

If that is so, then it is expected behaviour, unless you think that it worked with a previous version.

..Steve _______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev

Hi

First tests are good :).

Great!

But something strange happens with location of some house numbers. For example Pulawska 67 (in Warszawa city) is here: http://osm.org/go/0OyvXjAp5

but on Mapsource is shown here: http://osm.org/go/0OyvSy29r

This is more than 1 kilometer apart!

That would be odd, but I notice that in OSM one of the buildings is marked as number 13 (presumably in error - should be 113 is my guess). And 67 is pretty much mid way between 13 and 113. So could you check that as a source of the problem?

Also addresses like Municypalna 14 are shifted about 50m. Another bug

Numbers are spread evenly between road junctions, so you can't expect them to be exactly in position. We know how to deal with cases where houses are bunched on one part of a road between junctions which would allow us to improve that and it would be an interesting problem for someone to implement.

is related to uncommon house numbers like 12a, 67/69, etc. When given an address like Municypalna 12a, Mapsource tries to locate Municypalna 12

Searchable numbers must be just plain numbers (or if not, we don't know how to implement such things). What does 67/69 mean? It could be 67 and 69, 67, 68, 69, or building 67 appartment number 69 and interpretation might differ in different countries. Probably something that can be improved though.

(which is elsewhere, probably interpolated because no node with such housenumber is in OSM). When I enter address Pulawska 67/69, Mapsource shows Pulawska 623.

Main branch of mkgmap behaves in similar way, but addresses are shifted slightly different (When entering Municypalna 12a, Mapsource searches for Municypalna 23, Pulawska 67 is in different place - still wrong).

This doesn't seem to be related with --code-page switch.

Well, yes, I don't really expect that the work I am doing should have any affect on housenumber search that are not related to street names. Housenumber improvements are a separate task. Best wishes ..Steve