[mkgmap-dev] How to solve/debug weird problem

Fri Jan 21 19:32:41 GMT 2011

Hi Johannes,

On Thu, Jan 20, 2011 at 10:25:27PM +0100, Johannes Formann wrote:
>> --add-pois-to-areas --adjust-turn-headings --drive-on-right
>> --check-roundabouts --remove-short-arcs=3.3 --net --route --gmapsupp
>> --tdbfile --index M0001736.TYP
>>
>> You might try removing all of them first. If it does not crash, add
>> --route (which should imply --net).
>
>Starting without those options
>java -Xms256m -Xmx2560m -ea -jar mkgmap.jar --max-jobs=1 --latin1
>--code-page=1252 -c /home/osm/build/template.args

It's great that it crashes even without --route. I think that you should 
put back the --remove-short-arcs=3.3, so that you will get rid of the 
'zero length arc' warnings, maybe except the ones about splitting looped 
ways.

Now, what if you use a template.args where you remove all files after 
the problematic tile? Try to find out which tile it is. Your sample 
output mentions two different tile files around the crash: 59420055 and 
59420066. If you cannot modify the assertion message to display the file 
name, you can use 'binary search' on the template.args file, removing 
half the tiles until it no longer crashes.

Let me illustrate. Let's say you have tiles 59420001 through 59420100.  
First, remove all tiles between 59420001 and 59420050 in the 
template.args. If it still crashes, then try removing half of the 
remaining stuff: 59420075 through 59420100. If it does not crash, undo 
the last step and remove the other half, or remove a quarter, until you 
finally get a crash again. At the end, you should have a minimal input 
that reproduces the crash.

By the way, which mkgmap version are you using?

Unrelated to this mkgmap bug (or insufficient diagnostics), is there a 
reason why you are still using the osm.bz2 input? The osm.pbf is more 
compact and can be split quicker.

Best regards,

	Marko