[postgis-users] where to register to submit bug?

[Mark Cave-Ayland]

On Mon, 2007-10-29 at 23:25 +0100, ivan minčík wrote:
> I made pg_dump from the table. You can download it here:
>           http://disk.jabbim.cz/hrach@jabber.cz/bug_db.tar.gz
> 
> You have to:
>     1. change my database user (called ivo) to Your user
>     2. createdb somedb
>     3. psql -U youruser -f bug_db.sql somedb
> 
> try
>     select simplify(the_geom,100) from bug_table where icutjcpa = 
> '8601232244003';
> 
> 
> 
> is it ok that way?
> thanks
>     Ivan


Hi Ivan,

I finally found some time to look at this, and it turns out that what
you're seeing is a symptom of another bug in the geometry parser.

By enabling debug in the DP routines, I could see that the reason the
simplifier was crashing was that it was trying to simplify a polygon
consisting of a single point. Here is the text representation of your
polygon here:


bug=# select astext(the_geom) from bug_table where icutjcpa =
'8601232244003';
astext
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
 MULTIPOLYGON(((-333399.63 -1270249.25,-333416.78 -1270247.43,-333415.54
-1270231.88,-333411.49 -1270209.38)),((-333408.1 -1270235.32,-333408.8
-1270244.32,-333401.1 -1270245.02,-333400.5 -1270236.12,-333408.1
-1270235.32)),((-333397.27 -1270210.4)),((-333397.58 -1270214.37)))
(1 row)

bug=#


Now what is interesting is that the PostGIS geometry parser should
reject this input because your POLYGONs don't have closed rings. You can
see this if you try the following:


bug=# select astext('MULTIPOLYGON(((-333399.63 -1270249.25,-333416.78
-1270247.43,-333415.54 -1270231.88,-333411.49 -1270209.38)),((-333408.1
-1270235.32,-333408.8 -1270244.32,-333401.1 -1270245.02,-333400.5
-1270236.12,-333408.1 -1270235.32)),((-333397.27
-1270210.4)),((-333397.58 -1270214.37)))'::geometry); 
ERROR:  geometry contains non-closed rings 
bug=#


Which is the correct behaviour. So how on earth did these invalid
geometries get inserted into the database in the first place? Due to the
fact I was able to restore your dump, the finger points towards the
(E)WKB parser, and you can see this is true from the following:


bug=# select
astext('010600002011080000040000000103000000010000000400000052B81E855E5914C100000040E96133C1EC51B81EA35914C1E17A146EE76133C18FC2F5289E5914C114AE47E1D76133C15C8FC2F58D5914C114AE4761C16133C10103000000010000000500000066666666805914C11F85EB51DB6133C133333333835914C11F85EB51E46133C166666666645914C152B81E05E56133C100000000625914C1EC51B81EDC6133C166666666805914C11F85EB51DB6133C10103000000010000000100000048E17A14555914C166666666C26133C1010300000001000000010000001F85EB51565914C1EC51B85EC66133C1'::geometry);

astext 
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
 MULTIPOLYGON(((-333399.63 -1270249.25,-333416.78 -1270247.43,-333415.54
-1270231.88,-333411.49 -1270209.38)),((-333408.1 -1270235.32,-333408.8
-1270244.32,-333401.1 -1270245.02,-333400.5 -1270236.12,-333408.1
-1270235.32)),((-333397.27 -1270210.4)),((-333397.58 -1270214.37)))
(1 row)

bug=#


So the real bug in this case is that the (E)WKB parser doesn't seem to
attempt to validate any geometries being passed through it :(  I've had
a look at the parser code and it's quite a complicated beast with not
many comments, so I imagine a fix is going to a little time to come up
with.

However, for an immediate fix you need to either remove or correct the
invalid geometries. The following SQL should locate these geometries for
you:


bug=# select gid, icutjcpa, isvalid(the_geom) from bug_table where
isvalid(the_geom) = 'f';



HTH,

Mark.

-- 
ILande - Open Source Consultancy
http://www.ilande.co.uk